AI For College Students Free

AI For College Students Free — independent reviews, comparisons, pricing and step-by-step guides on Aizhi.

• 

  Software engine

  A software engine is a core component of a complex software system. The word "engine" is a metaphor of a car's engine. Thus a software engine is a complex subsystem; not unlike how a car engine functions. Software engines work in conjunction with other components of a process or system. They typically have an input and an output, and the productivity is usually linear to running speed. There is no formal guideline for what should be called an engine, but the term has become widespread in the software industry. == Notable examples == === Multi-engine systems === Mainstream web browsers have both a browser engine and a JavaScript engine. Video games are often based on a game engine. Some of these also have specialized physics or graphics engines.

  Read more →
• 

  Contrastive Language-Image Pre-training

  Contrastive Language-Image Pre-training (CLIP) is a technique for training a pair of neural network models, one for image understanding and one for text understanding, using a contrastive objective. This method has enabled broad applications across multiple domains, including cross-modal retrieval, text-to-image generation, and aesthetic ranking. == Algorithm == The CLIP method trains a pair of models contrastively. One model takes in a piece of text as input and outputs a single vector representing its semantic content. The other model takes in an image and similarly outputs a single vector representing its visual content. The models are trained so that the vectors corresponding to semantically similar text-image pairs are close together in the shared vector space, while those corresponding to dissimilar pairs are far apart. To train a pair of CLIP models, one would start by preparing a large dataset of image-caption pairs. During training, the models are presented with batches of N {\displaystyle N} image-caption pairs. Let the outputs from the text and image models be respectively v 1 , . . . , v N , w 1 , . . . , w N {\displaystyle v_{1},...,v_{N},w_{1},...,w_{N}} . Two vectors are considered "similar" if their dot product is large. The loss incurred on this batch is the multi-class N-pair loss, which is a symmetric cross-entropy loss over similarity scores: − 1 N ∑ i ln ⁡ e v i ⋅ w i / T ∑ j e v i ⋅ w j / T − 1 N ∑ j ln ⁡ e v j ⋅ w j / T ∑ i e v i ⋅ w j / T {\displaystyle -{\frac {1}{N}}\sum _{i}\ln {\frac {e^{v_{i}\cdot w_{i}/T}}{\sum _{j}e^{v_{i}\cdot w_{j}/T}}}-{\frac {1}{N}}\sum _{j}\ln {\frac {e^{v_{j}\cdot w_{j}/T}}{\sum _{i}e^{v_{i}\cdot w_{j}/T}}}} In essence, this loss function encourages the dot product between matching image and text vectors ( v i ⋅ w i {\displaystyle v_{i}\cdot w_{i}} ) to be high, while discouraging high dot products between non-matching pairs. The parameter T > 0 {\displaystyle T>0} is the temperature, which is parameterized in the original CLIP model as T = e − τ {\displaystyle T=e^{-\tau }} where τ ∈ R {\displaystyle \tau \in \mathbb {R} } is a learned parameter. Other loss functions are possible. For example, Sigmoid CLIP (SigLIP) proposes the following loss function: L = 1 N ∑ i , j ∈ 1 : N f ( ( 2 δ i , j − 1 ) ( e τ w i ⋅ v j + b ) ) {\displaystyle L={\frac {1}{N}}\sum _{i,j\in 1:N}f((2\delta _{i,j}-1)(e^{\tau }w_{i}\cdot v_{j}+b))} where f ( x ) = ln ⁡ ( 1 + e − x ) {\displaystyle f(x)=\ln(1+e^{-x})} is the negative log sigmoid loss, and the Dirac delta symbol δ i , j {\displaystyle \delta _{i,j}} is 1 if i = j {\displaystyle i=j} else 0. == CLIP models == While the original model was developed by OpenAI, subsequent models have been trained by other organizations as well. === Image model === The image encoding models used in CLIP are typically vision transformers (ViT). The naming convention for these models often reflects the specific ViT architecture used. For instance, "ViT-L/14" means a "vision transformer large" (compared to other models in the same series) with a patch size of 14, meaning that the image is divided into 14-by-14 pixel patches before being processed by the transformer. The size indicator ranges from B, L, H, G (base, large, huge, giant), in that order. Other than ViT, the image model is typically a convolutional neural network, such as ResNet (in the original series by OpenAI), or ConvNeXt (in the OpenCLIP model series by LAION). Since the output vectors of the image model and the text model must have exactly the same length, both the image model and the text model have fixed-length vector outputs, which in the original report is called "embedding dimension". For example, in the original OpenAI model, the ResNet models have embedding dimensions ranging from 512 to 1024, and for the ViTs, from 512 to 768. Its implementation of ViT was the same as the original one, with one modification: after position embeddings are added to the initial patch embeddings, there is a LayerNorm. Its implementation of ResNet was the same as the original one, with 3 modifications: In the start of the CNN (the "stem"), they used three stacked 3x3 convolutions instead of a single 7x7 convolution, as suggested by. There is an average pooling of stride 2 at the start of each downsampling convolutional layer (they called it rect-2 blur pooling according to the terminology of ). This has the effect of blurring images before downsampling, for antialiasing. The final convolutional layer is followed by a multiheaded attention pooling. ALIGN a model with similar capabilities, trained by researchers from Google used EfficientNet, a kind of convolutional neural network. === Text model === The text encoding models used in CLIP are typically Transformers. In the original OpenAI report, they reported using a Transformer (63M-parameter, 12-layer, 512-wide, 8 attention heads) with lower-cased byte pair encoding (BPE) with 49152 vocabulary size. Context length was capped at 76 for efficiency. Like GPT, it was decoder-only, with only causally-masked self-attention. Its architecture is the same as GPT-2. Like BERT, the text sequence is bracketed by two special tokens [SOS] and [EOS] ("start of sequence" and "end of sequence"). Take the activations of the highest layer of the transformer on the [EOS], apply LayerNorm, then a final linear map. This is the text encoding of the input sequence. The final linear map has output dimension equal to the embedding dimension of whatever image encoder it is paired with. These models all had context length 77 and vocabulary size 49408. ALIGN used BERT of various sizes. == Dataset == === WebImageText === The CLIP models released by OpenAI were trained on a dataset called "WebImageText" (WIT) containing 400 million pairs of images and their corresponding captions scraped from the internet. The total number of words in this dataset is similar in scale to the WebText dataset used for training GPT-2, which contains about 40 gigabytes of text data. The dataset contains 500,000 text-queries, with up to 20,000 (image, text) pairs per query. The text-queries were generated by starting with all words occurring at least 100 times in English Wikipedia, then extended by bigrams with high mutual information, names of all Wikipedia articles above a certain search volume, and WordNet synsets. The dataset is private and has not been released to the public, and there is no further information on it. ==== Data preprocessing ==== For the CLIP image models, the input images are preprocessed by first dividing each of the R, G, B values of an image by the maximum possible value, so that these values fall between 0 and 1, then subtracting by [0.48145466, 0.4578275, 0.40821073], and dividing by [0.26862954, 0.26130258, 0.27577711]. The rationale was that these are the mean and standard deviations of the images in the WebImageText dataset, so this preprocessing step roughly whitens the image tensor. These numbers slightly differ from the standard preprocessing for ImageNet, which uses [0.485, 0.456, 0.406] and [0.229, 0.224, 0.225]. If the input image does not have the same resolution as the native resolution (224×224 for all except ViT-L/14@336px, which has 336×336 resolution), then the input image is first scaled by bicubic interpolation, so that its shorter side is the same as the native resolution, then the central square of the image is cropped out. === Others === ALIGN used over one billion image-text pairs, obtained by extracting images and their alt-tags from online crawling. The method was described as similar to how the Conceptual Captions dataset was constructed, but instead of complex filtering, they only applied a frequency-based filtering. Later models trained by other organizations had published datasets. For example, LAION trained OpenCLIP with published datasets LAION-400M, LAION-2B, and DataComp-1B. == Training == In the original OpenAI CLIP report, they reported training 5 ResNet and 3 ViT (ViT-B/32, ViT-B/16, ViT-L/14). Each was trained for 32 epochs. The largest ResNet model took 18 days to train on 592 V100 GPUs. The largest ViT model took 12 days on 256 V100 GPUs. All ViT models were trained on 224×224 image resolution. The ViT-L/14 was then boosted to 336×336 resolution by FixRes, resulting in a model. They found this was the best-performing model. In the OpenCLIP series, the ViT-L/14 model was trained on 384 A100 GPUs on the LAION-2B dataset, for 160 epochs for a total of 32B samples seen. == Applications == === Cross-modal retrieval === CLIP's cross-modal retrieval enables the alignment of visual and textual data in a shared latent space, allowing users to retrieve images based on text descriptions and vice versa, without the need for explicit image annotations. In text-to-image retrieval, users input descriptive text, and CLIP retrieves images with matching embeddings. In image-to-text retrieval, images are used to find related text content. CLIP’s ability to connect vis

  Read more →
• 

  Embodied cognitive science

  Embodied cognitive science is an interdisciplinary field of research, the aim of which is to explain the mechanisms underlying intelligent behavior. It comprises three main methodologies: the modeling of psychological and biological systems in a holistic manner that considers the mind and body as a single entity; the formation of a common set of general principles of intelligent behavior; and the experimental use of robotic agents in controlled environments. == Contributors == Embodied cognitive science borrows heavily from embodied philosophy and the related research fields of cognitive science, psychology, neuroscience and artificial intelligence. Contributors to the field include: From the perspective of neuroscience, Gerald Edelman of the Neurosciences Institute at La Jolla, Francisco Varela of CNRS in France, and J. A. Scott Kelso of Florida Atlantic University From the perspective of psychology, Lawrence Barsalou, Michael Turvey, Vittorio Guidano and Eleanor Rosch From the perspective of linguistics, Gilles Fauconnier, George Lakoff, Mark Johnson, Leonard Talmy and Mark Turner From the perspective of language acquisition, Eric Lenneberg and Philip Rubin at Haskins Laboratories From the perspective of anthropology, Edwin Hutchins, Bradd Shore, James Wertsch and Merlin Donald. From the perspective of autonomous agent design, early work is sometimes attributed to Rodney Brooks or Valentino Braitenberg From the perspective of artificial intelligence, Understanding Intelligence by Rolf Pfeifer and Christian Scheier or How the Body Shapes the Way We Think, by Rolf Pfeifer and Josh C. Bongard From the perspective of philosophy, Andy Clark, Dan Zahavi, Shaun Gallagher, and Evan Thompson In 1950, Alan Turing proposed that a machine may need a human-like body to think and speak: It can also be maintained that it is best to provide the machine with the best sense organs that money can buy, and then teach it to understand and speak English. That process could follow the normal teaching of a child. Things would be pointed out and named, etc. Again, I do not know what the right answer is, but I think both approaches should be tried. == Traditional cognitive theory == Embodied cognitive science is an alternative theory to cognition in which it minimizes appeals to computational theory of mind in favor of greater emphasis on how an organism's body determines how and what it thinks. Traditional cognitive theory is based mainly around symbol manipulation, in which certain inputs are fed into a processing unit that produces an output. These inputs follow certain rules of syntax, from which the processing unit finds semantic meaning. Thus, an appropriate output is produced. For example, a human's sensory organs are its input devices, and the stimuli obtained from the external environment are fed into the nervous system which serves as the processing unit. From here, the nervous system is able to read the sensory information because it follows a syntactic structure, thus an output is created. This output then creates bodily motions and brings forth behavior and cognition. Of particular note is that cognition is sealed away in the brain, meaning that mental cognition is cut off from the external world and is only possible by the input of sensory information. == The embodied cognitive approach == Embodied cognitive science differs from the traditionalist approach in that it denies the input-output system. This is chiefly due to the problems presented by the Homunculus argument, which concluded that semantic meaning could not be derived from symbols without some kind of inner interpretation. If some little man in a person's head interpreted incoming symbols, then who would interpret the little man's inputs? Because of the specter of an infinite regress, the traditionalist model began to seem less plausible. Thus, embodied cognitive science aims to avoid this problem by defining cognition in three ways. === Physical attributes of the body === The first aspect of embodied cognition examines the role of the physical body, particularly how its properties affect its ability to think. This part attempts to overcome the symbol manipulation component that is a feature of the traditionalist model. Depth perception, for instance, can be better explained under the embodied approach due to the sheer complexity of the action. Depth perception requires that the brain detect the disparate retinal images obtained by the distance of the two eyes. In addition, body and head cues complicate this further. When the head is turned in a given direction, objects in the foreground will appear to move against objects in the background. From this, it is said that some kind of visual processing is occurring without the need of any kind of symbol manipulation. This is because the objects appearing to move the foreground are simply appearing to move. This observation concludes then that depth can be perceived with no intermediate symbol manipulation necessary. A more poignant example exists through examining auditory perception. Generally speaking the greater the distance between the ears, the greater the possible auditory acuity. Also relevant is the amount of density in between the ears, for the strength of the frequency wave alters as it passes through a given medium. The brain's auditory system takes these factors into account as it process information, but again without any need for a symbolic manipulation system. This is because the distance between the ears for example does not need symbols to represent it. The distance itself creates the necessary opportunity for greater auditory acuity. The amount of density between the ears is similar, in that it is the actual amount itself that simply forms the opportunity for frequency alteration. Thus under consideration of the physical properties of the body, a symbolic system is unnecessary and an unhelpful metaphor. === The body's role in the cognitive process === The second aspect draws heavily from George Lakoff's and Mark Johnson's work on concepts. They argued that humans use metaphors whenever possible to better explain their external world. Humans also have a basic stock of concepts in which other concepts can be derived from. These basic concepts include spatial orientations such as up, down, front, and back. Humans can understand what these concepts mean because they can directly experience them from their own bodies. For example, because human movement revolves around standing erect and moving the body in an up-down motion, humans innately have these concepts of up and down. Lakoff and Johnson contend this is similar with other spatial orientations such as front and back too. As mentioned earlier, these basic stocks of spatial concepts are the basis in which other concepts are constructed. Happy and sad for instance are seen now as being up or down respectively. When someone says they are feeling down, what they are really saying is that they feel sad for example. Thus the point here is that true understanding of these concepts is contingent on whether one can have an understanding of the human body. So the argument goes that if one lacked a human body, they could not possibly know what up or down could mean, or how it could relate to emotional states. [I]magine a spherical being living outside of any gravitational field, with no knowledge or imagination of any other kind of experience. What could UP possibly mean to such a being? While this does not mean that such beings would be incapable of expressing emotions in other words, it does mean that they would express emotions differently from humans. Human concepts of happiness and sadness would be different because human would have different bodies. So then an organism's body directly affects how it can think, because it uses metaphors related to its body as the basis of concepts. === Interaction of local environment === A third component of the embodied approach looks at how agents use their immediate environment in cognitive processing. Meaning, the local environment is seen as an actual extension of the body's cognitive process. The example of a personal digital assistant (PDA) is used to better imagine this. Echoing functionalism (philosophy of mind), this point claims that mental states are individuated by their role in a much larger system. So under this premise, the information on a PDA is similar to the information stored in the brain. So then if one thinks information in the brain constitutes mental states, then it must follow that information in the PDA is a cognitive state too. Consider also the role of pen and paper in a complex multiplication problem. The pen and paper are so involved in the cognitive process of solving the problem that it seems ridiculous to say they are somehow different from the process, in very much the same way the PDA is used for information like the brain. Another example examines how humans control and manipulate their environment

  Read more →
• 

  Hierarchical control system

  A hierarchical control system (HCS) is a form of control system in which a set of devices and governing software is arranged in a hierarchical tree. When the links in the tree are implemented by a computer network, then that hierarchical control system is also a form of networked control system. == Overview == A human-built system with complex behavior is often organized as a hierarchy. For example, a command hierarchy has among its notable features the organizational chart of superiors, subordinates, and lines of organizational communication. Hierarchical control systems are organized similarly to divide the decision making responsibility. Each element of the hierarchy is a linked node in the tree. Commands, tasks and goals to be achieved flow down the tree from superior nodes to subordinate nodes, whereas sensations and command results flow up the tree from subordinate to superior nodes. Nodes may also exchange messages with their siblings. The two distinguishing features of a hierarchical control system are related to its layers. Each higher layer of the tree operates with a longer interval of planning and execution time than its immediately lower layer. The lower layers have local tasks, goals, and sensations, and their activities are planned and coordinated by higher layers which do not generally override their decisions. The layers form a hybrid intelligent system in which the lowest, reactive layers are sub-symbolic. The higher layers, having relaxed time constraints, are capable of reasoning from an abstract world model and performing planning. A hierarchical task network is a good fit for planning in a hierarchical control system. Besides artificial systems, an animal's control systems are proposed to be organized as a hierarchy. In perceptual control theory, which postulates that an organism's behavior is a means of controlling its perceptions, the organism's control systems are suggested to be organized in a hierarchical pattern as their perceptions are constructed so. == Control system structure == The accompanying diagram is a general hierarchical model which shows functional manufacturing levels using computerised control of an industrial control system. Referring to the diagram; Level 0 contains the field devices such as flow and temperature sensors, and final control elements, such as control valves Level 1 contains the industrialised Input/Output (I/O) modules, and their associated distributed electronic processors. Level 2 contains the supervisory computers, which collate information from processor nodes on the system, and provide the operator control screens. Level 3 is the production control level, which does not directly control the process, but is concerned with monitoring production and monitoring targets Level 4 is the production scheduling level. == Applications == === Manufacturing, robotics and vehicles === Among the robotic paradigms is the hierarchical paradigm in which a robot operates in a top-down fashion, heavy on planning, especially motion planning. Computer-aided production engineering has been a research focus at NIST since the 1980s. Its Automated Manufacturing Research Facility was used to develop a five layer production control model. In the early 1990s DARPA sponsored research to develop distributed (i.e. networked) intelligent control systems for applications such as military command and control systems. NIST built on earlier research to develop its Real-Time Control System (RCS) and Real-time Control System Software which is a generic hierarchical control system that has been used to operate a manufacturing cell, a robot crane, and an automated vehicle. In November 2007, DARPA held the Urban Challenge. The winning entry, Tartan Racing employed a hierarchical control system, with layered mission planning, motion planning, behavior generation, perception, world modelling, and mechatronics. === Artificial intelligence === Subsumption architecture is a methodology for developing artificial intelligence that is heavily associated with behavior based robotics. This architecture is a way of decomposing complicated intelligent behavior into many "simple" behavior modules, which are in turn organized into layers. Each layer implements a particular goal of the software agent (i.e. system as a whole), and higher layers are increasingly more abstract. Each layer's goal subsumes that of the underlying layers, e.g. the decision to move forward by the eat-food layer takes into account the decision of the lowest obstacle-avoidance layer. Behavior need not be planned by a superior layer, rather behaviors may be triggered by sensory inputs and so are only active under circumstances where they might be appropriate. Reinforcement learning has been used to acquire behavior in a hierarchical control system in which each node can learn to improve its behavior with experience. James Albus, while at NIST, developed a theory for intelligent system design named the Reference Model Architecture (RMA), which is a hierarchical control system inspired by RCS. Albus defines each node to contain these components. Behavior generation is responsible for executing tasks received from the superior, parent node. It also plans for, and issues tasks to, the subordinate nodes. Sensory perception is responsible for receiving sensations from the subordinate nodes, then grouping, filtering, and otherwise processing them into higher level abstractions that update the local state and which form sensations that are sent to the superior node. Value judgment is responsible for evaluating the updated situation and evaluating alternative plans. World Model is the local state that provides a model for the controlled system, controlled process, or environment at the abstraction level of the subordinate nodes. At its lowest levels, the RMA can be implemented as a subsumption architecture, in which the world model is mapped directly to the controlled process or real world, avoiding the need for a mathematical abstraction, and in which time-constrained reactive planning can be implemented as a finite-state machine. Higher levels of the RMA however, may have sophisticated mathematical world models and behavior implemented by automated planning and scheduling. Planning is required when certain behaviors cannot be triggered by current sensations, but rather by predicted or anticipated sensations, especially those that come about as result of the node's actions.

  Read more →
• 

  Weight initialization

  In deep learning, weight initialization or parameter initialization describes the initial step in creating a neural network. A neural network contains trainable parameters that are modified during training: weight initialization is the pre-training step of assigning initial values to these parameters. The choice of weight initialization method affects the speed of convergence, the scale of neural activation within the network, the scale of gradient signals during backpropagation, and the quality of the final model. Proper initialization is necessary for avoiding issues such as vanishing and exploding gradients and activation function saturation. Note that even though this article is titled "weight initialization", both weights and biases are used in a neural network as trainable parameters, so this article describes how both of these are initialized. Similarly, trainable parameters in convolutional neural networks (CNNs) are called kernels and biases, and this article also describes these. == Constant initialization == We discuss the main methods of initialization in the context of a multilayer perceptron (MLP). Specific strategies for initializing other network architectures are discussed in later sections. For an MLP, there are only two kinds of trainable parameters, called weights and biases. Each layer l {\displaystyle l} contains a weight matrix W ( l ) ∈ R n l − 1 × n l {\displaystyle W^{(l)}\in \mathbb {R} ^{n_{l-1}\times n_{l}}} and a bias vector b ( l ) ∈ R n l {\displaystyle b^{(l)}\in \mathbb {R} ^{n_{l}}} , where n l {\displaystyle n_{l}} is the number of neurons in that layer. A weight initialization method is an algorithm for setting the initial values for W ( l ) , b ( l ) {\displaystyle W^{(l)},b^{(l)}} for each layer l {\displaystyle l} . The simplest form is zero initialization: W ( l ) = 0 , b ( l ) = 0 {\displaystyle W^{(l)}=0,b^{(l)}=0} Zero initialization is usually used for initializing biases, but it is not used for initializing weights, as it leads to symmetry in the network, causing all neurons to learn the same features. In this page, we assume b = 0 {\displaystyle b=0} unless otherwise stated. Recurrent neural networks typically use activation functions with bounded range, such as sigmoid and tanh, since unbounded activation may cause exploding values. (Le, Jaitly, Hinton, 2015) suggested initializing weights in the recurrent parts of the network to identity and zero bias, similar to the idea of residual connections and LSTM with no forget gate. In most cases, the biases are initialized to zero, though some situations can use a nonzero initialization. For example, in multiplicative units, such as the forget gate of LSTM, the bias can be initialized to 1 to allow good gradient signal through the gate. For neurons with ReLU activation, one can initialize the bias to a small positive value like 0.1, so that the gradient is likely nonzero at initialization, avoiding the dying ReLU problem. == Random initialization == Random initialization means sampling the weights from a normal distribution or a uniform distribution, usually independently. === LeCun initialization === LeCun initialization, popularized in (LeCun et al., 1998), is designed to preserve the variance of neural activations during the forward pass. It samples each entry in W ( l ) {\displaystyle W^{(l)}} independently from a distribution with mean 0 and variance 1 / n l − 1 {\displaystyle 1/n_{l-1}} . For example, if the distribution is a continuous uniform distribution, then the distribution is U ( ± 3 / n l − 1 ) {\displaystyle {\mathcal {U}}(\pm {\sqrt {3/n_{l-1}}})} . === Glorot initialization === Glorot initialization (or Xavier initialization) was proposed by Xavier Glorot and Yoshua Bengio. It was designed as a compromise between two goals: to preserve activation variance during the forward pass and to preserve gradient variance during the backward pass. For uniform initialization, it samples each entry in W ( l ) {\displaystyle W^{(l)}} independently and identically from U ( ± 6 / ( n l + 1 + n l − 1 ) ) {\displaystyle {\mathcal {U}}(\pm {\sqrt {6/(n_{l+1}+n_{l-1})}})} . In the context, n l − 1 {\displaystyle n_{l-1}} is also called the "fan-in", and n l + 1 {\displaystyle n_{l+1}} the "fan-out". When the fan-in and fan-out are equal, then Glorot initialization is the same as LeCun initialization. === He initialization === As Glorot initialization performs poorly for ReLU activation, He initialization (or Kaiming initialization) was proposed by Kaiming He et al. for networks with ReLU activation. It samples each entry in W ( l ) {\displaystyle W^{(l)}} from N ( 0 , 2 / n l − 1 ) {\displaystyle {\mathcal {N}}(0,2/n_{l-1})} . === Orthogonal initialization === (Saxe et al. 2013) proposed orthogonal initialization: initializing weight matrices as uniformly random (according to the Haar measure) semi-orthogonal matrices, multiplied by a factor that depends on the activation function of the layer. It was designed so that if one initializes a deep linear network this way, then its training time until convergence is independent of depth. Sampling a uniformly random semi-orthogonal matrix can be done by initializing X {\displaystyle X} by IID sampling its entries from a standard normal distribution, then calculate ( X X ⊤ ) − 1 / 2 X {\displaystyle \left(XX^{\top }\right)^{-1/2}X} or its transpose, depending on whether X {\displaystyle X} is tall or wide. For CNN kernels with odd widths and heights, orthogonal initialization is done this way: initialize the central point by a semi-orthogonal matrix, and fill the other entries with zero. As an illustration, a kernel K {\displaystyle K} of shape 3 × 3 × c × c ′ {\displaystyle 3\times 3\times c\times c'} is initialized by filling K [ 2 , 2 , : , : ] {\displaystyle K[2,2,:,:]} with the entries of a random semi-orthogonal matrix of shape c × c ′ {\displaystyle c\times c'} , and the other entries with zero. (Balduzzi et al., 2017) used it with stride 1 and zero-padding. This is sometimes called the Orthogonal Delta initialization. Related to this approach, unitary initialization proposes to parameterize the weight matrices to be unitary matrices, with the result that at initialization they are random unitary matrices (and throughout training, they remain unitary). This is found to improve long-sequence modelling in LSTM. Orthogonal initialization has been generalized to layer-sequential unit-variance (LSUV) initialization. It is a data-dependent initialization method, and can be used in convolutional neural networks. It first initializes weights of each convolution or fully connected layer with orthonormal matrices. Then, proceeding from the first to the last layer, it runs a forward pass on a random minibatch, and divides the layer's weights by the standard deviation of its output, so that its output has variance approximately 1. === Fixup initialization === In 2015, the introduction of residual connections allowed very deep neural networks to be trained, much deeper than the ~20 layers of the previous state of the art (such as the VGG-19). Residual connections gave rise to their own weight initialization problems and strategies. These are sometimes called "normalization-free" methods, since using residual connection could stabilize the training of a deep neural network so much that normalizations become unnecessary. Fixup initialization is designed specifically for networks with residual connections and without batch normalization, as follows: Initialize the classification layer and the last layer of each residual branch to 0. Initialize every other layer using a standard method (such as He initialization), and scale only the weight layers inside residual branches by L − 1 2 m − 2 {\displaystyle L^{-{\frac {1}{2m-2}}}} . Add a scalar multiplier (initialized at 1) in every branch and a scalar bias (initialized at 0) before each convolution, linear, and element-wise activation layer. Similarly, T-Fixup initialization is designed for Transformers without layer normalization. === Others === Instead of initializing all weights with random values on the order of O ( 1 / n ) {\displaystyle O(1/{\sqrt {n}})} , sparse initialization initialized only a small subset of the weights with larger random values, and the other weights zero, so that the total variance is still on the order of O ( 1 ) {\displaystyle O(1)} . Random walk initialization was designed for MLP so that during backpropagation, the L2 norm of gradient at each layer performs an unbiased random walk as one moves from the last layer to the first. Looks linear initialization was designed to allow the neural network to behave like a deep linear network at initialization, since W R e L U ( x ) − W R e L U ( − x ) = W x {\displaystyle W\;\mathrm {ReLU} (x)-W\;\mathrm {ReLU} (-x)=Wx} . It initializes a matrix W {\displaystyle W} of shape R n 2 × m {\displaystyle \mathbb {R} ^{{\frac {n}{2}}\times m}} by any method, such as orthogonal initialization, t

  Read more →
• 

  Nolot

  Nolot is a chess test suite with 11 positions from real games. They were compiled by Pierre Nolot (French: [nɔ.lo]) for the French chess magazine Gambisco and posted on the rec.games.chess Usenet group in 1994. They were designed to be particularly hard to solve for chess engines to solve at the time, although modern engines can find a solution near-instantaneously. == Problem 1 == FEN: r3qb1k/1b4p1/p2pr2p/3n4/Pnp1N1N1/6RP/1B3PP1/1B1QR1K1 w - - 0 1 26.Nxh6!! c3 (26... Rxh6 27.Nxd6 Qh5 (best) 28.Rg5! Qxd1 29.Nf7+ Kg8 30.Nxh6+ Kh8 31.Rxd1 c3 32.Nf7+ Kg8 33.Bg6! Nf4 34.Bxc3 Nxg6 35.Bxb4 Kxf7 36.Rd7+ Kf6 37.Rxg6+ Kxg6 38.Rxb7 ±) 27.Nf5! cxb2 28.Qg4 Bc8 (28... g6!? 29.Kh2! 29.Qd7 30.Nh4 Bc6 31.Nc5! dxc 32.Rxe6 Nf6 33.Nxg6+ Kg7 34.Qg5 Nbd5 35.Ne5 Kh8 36.Nxd7 ±) 29.Qh4+ Rh6 30.Nxh6 gxh6 31.Kh2! Qe5 32.Ng5 Qf6 33.Re8 Bf5 34.Qxh6 (missing a mate in 6: 34.Nf7+ Qxf7 35.Qxh6+ Bh7 36.Rxa8 Nf6 37.Rxf8 Qxf8 38.Qxf8+ Ng8 39.Qg7#) 34...Qxh6 35.Nf7+ Kh7 36.Bxf5+ Qg6 37.Bxg6+ Kg7 38.Rxa8 Be7 39.Rb8 a5 40.Be4+ Kxf7 41.Bxd5+ 1–0 The best Novag computer, the Diablo 68000, finds 26. Nxh6 after seven and a half months (Pierre Nolot has let it run on the position for 14 months and one day, until a power failure stopped an analysis of over 80,000,000,000 nodes.) but for wrong reasons: it evaluates white's position as inferior and thinks this move would enable it to draw. Today Gambit Tiger 2.0 for example can find it quite quickly: Most free engines running on 64-bit processors in 2010 could solve this problem and the others in a few seconds. 1.Qd4 c3 2.Bxc3 Nxc3 3.Qxb4 Nxe4 4.Qxb7 Rb8 5.Qxb8 Qxb8 6.Bxe4 d5 7.Rb1 μ (-1.20) Depth: 12 00:00:09 6055 kN 1.Nxh6 c3 2.Nf5 cxb2 3.Qg4 Rb8 4.Nxg7 Rg6 5.Qxg6 Qxg6 6.Rxg6 Bxg7 7.Nxd6 ³ (-0.48) Depth: 12 00:00:21 14368 kN 1.Nxh6 c3 2.Nf5 cxb2 3.Qg4 Rc8 4.Nxg7 Rg6 5.Nxe8 Rxg4 6.Rxg4 Rxe8 7.Rg6 μ (-0.74) Depth: 13 00:00:55 38455 kN 1.Ne3 Rxe4 2.Bxe4 Qxe4 3.Nxd5 Qxd5 4.Qc1 Qf5 5.Qxh6+ Qh7 6.Qe6 Nd3 7.Re2 Nxb2 8.Rxb2 ³ (-0.58) Depth: 13 00:01:30 62979 kN 1.Ne3 Rxe4 ³ (-0.58) Depth: 14 00:02:02 84941 kN 1.Ne3 Nxe3 2.Rexe3 Bxe4 3.Qg4 Rg6 4.Qxe4 Qxe4 5.Bxe4 Rxg3 6.Rxg3 d5 7.Bf5 Re8 8.Bc3 ³ (-0.30) Depth: 15 00:03:05 128968 kN 1.Nxh6 ² (0.32) Depth: 15 00:07:58 350813 kN With the next ply showing a clear advantage. Stockfish 14dev 64bit 4CPU running on 2020 hardware recognises the significance of Nxh6!! in 1 second. Stockfish_21092606_x64_avx2: NNUE evaluation using nn-13406b1dcbe0.nnue enabled. 19/32 00:01 7708k 4882k +3,00 Nxh6 Rxh6 Nxd6 Qh5 Bg6 Qxd1 Nf7+ Kg8 Nxh6+ gxh6 Bh5+ Kh7 Rxd1 c3 Bxc3 Nxc3 Rd7+ Kh8 Rxb7 Ne4 Re3 Nxf2 Kxf2 Bc5 Ke2 Bxe3 Kxe3 Nd5+ Kf2 49/73 15:02 5118270k 5673k +6,15 Nxh6 Rxh6 Nxd6 Qh5 Rg5 Qxd1 Nf7+ Kg8 Nxh6+ Kh8 Rxd1 c3 Nf7+ Kg8 Bg6 Nf4 Bxc3 Nbd5 Rb1 Bc6 Bd2 Nxg6 Rxg6 Ne7 Rxc6 Nxc6 Rb6 Rc8 Ng5 a5 Ra6 Bb4 Be3 Ne5 Bd4 Nc6 Bb6 Bd2 h4 Kf8 Bc5+ Kg8 Be3 Bxe3 fxe3 Kf8 Kf2 Ke7 Nf3 Kd7 Rb6 Ne7 Rb5 Kd6 Rxa5 Rc2+ Kg3 Re2 Nd4 Rxe3+ Kf4 Rd3 Nf5+ Kc7 Nxe7 == Problem 2 == FEN: r4rk1/pp1n1p1p/1nqP2p1/2b1P1B1/4NQ2/1B3P2/PP2K2P/2R5 w - - 0 1 22.Rxc5!! Nxc5 23.Nf6+ Kh8 24.Qh4 Qb5+ (computers think there is perpetual check here, but...) 25.Ke3! 25... h5 26.Nxh5 Qxb3+ (26... d5+ 27.Bxd5 Qd3 28.Kf2 Ne4+ 29.Bxe4 Qd4+ 30.Kg2 Qxb2+ 31.Kh3 ±) and White won in 41 moves. Today Deep Junior 8.ZX for example finds it very quickly (around 1 minute): 1.Kd1 Rac8 2.Bh6 Qb5 3.Rc3 Qf1+ 4.Kc2 Rc6 5.Bxf8 −+ (-2.11) Depth: 12 00:00:04 10422 kN 1.Nxc5 Nxc5 2.Rxc5 Qxc5 3.e6 Rae8 4.e7 Nc8 5.Kf1 Nxd6 6.Bf6 b5 −+ (-2.10) Depth: 12 00:00:14 25054 kN 1.Bf6! μ (-1.35) Depth: 12 00:00:17 34601 kN 1.Bf6 Qb5+ 2.Ke1 Bb4+ 3.Kf2 Bc5+ = (0.00) Depth: 12 00:00:20 34601 kN 1.Bf6 Qb5+ 2.Ke1 Nxf6 3.Nxf6+ Kg7 4.Nh5+ gxh5 5.Qf6+ Kg8 6.Qg5+ Kh8 7.Qf6+ = (0.00) Depth: 15 00:01:01 130544 kN 1.Rxc5! = (0.15) Depth: 15 00:01:12 145875 kN 1.Rxc5 Nxc5 2.Nf6+ Kh8 3.Qh4 Qb5+ 4.Ke3 h5 5.Nxh5 Qd3+ 6.Kf2 Ne4+ 7.fxe4 Qd4+ 8.Kf1 Qd3+ 9.Ke1 Qb1+ 10.Bd1 ± (2.18) Depth: 15 00:01:18 145875 kN Stockfish 14dev 64bit 4CPU running on 2020 hardware recognises the significance of Rxc5!! in 1 second. Stockfish_21092606_x64_avx2: NNUE evaluation using nn-13406b1dcbe0.nnue enabled. 21/25 00:01 5822k 5545k +6,61 Rxc5 Qxc5 Nxc5 Nxc5 Bh6 Nbd7 Bxf8 Rxf8 Qe3 Rc8 f4 Nxe5 Qxe5 Ne6 Bxe6 Rc2+ Kd3 Rxh2 46/86 11:27 5057055k 7355k +7,61 Rxc5 Qxc5 Nxc5 Nxc5 Bf6 Ne6 Qh6 Nd4+ Kf2 Nf5 Qg5 Nd7 h4 Nxf6 Qxf6 Ng7 d7 b5 Bd5 Rab8 b4 Nh5 Bxf7+ Rxf7 d8R+ Rxd8 Qxd8+ Rf8 Qd5+ Kg7 e6 Kf6 Qd7 Ng7 Qd4+ Kxe6 Qxg7 Rf7 Qc3 Ke7 Qc5+ Ke8 Qc8+ Ke7 h5 gxh5 Kg3 h4+ Kh2 h6 Qc5+ Kf6 Qxb5 Kg7 f4 Rxf4 Qe5+ Rf6 b5 h3 Qd4 Kg8 Qxf6 h5 Blacks 22. .. Nxc5 is suboptimal and leads faster mate 77/44 09:18 6987714k 12518k +M22 Nf6+ Kh8 Qh4 Qb5+ Ke3 Qxb3+ axb3 h5 Nxh5 Nd5+ Kd4 Ne6+ Kxd5 Nxg5 Qxg5 gxh5 f4 Rad8 f5 f6 Qxh5+ Kg7 Qg6+ Kh8 e6 b6 e7 Rb8 exf8Q+ Rxf8 Ke6 b5 Ke7 Rb8 Qh5+ Kg7 Qf7+ Kh8 Kxf6 Rf8 Qxf8+ Kh7 Qg7+ == Problem 3 == FEN: r2qk2r/ppp1b1pp/2n1p3/3pP1n1/3P2b1/2PB1NN1/PP4PP/R1BQK2R w KQkq - 0 1 12.Nxg5!! Bxd1 13.Nxe6 Qb8 14.Nxg7+!! Kf8 15.Bh6! Bg4 16.0-0+ Kg8 17.Rf4 ± White wins with a queen sac but black has defensive resources. Stockfish 8 64bit 3CPU running on 2016 hardware recognizes the significance of Nxg5!! in 55 seconds. Stockfish 14 dev (Stockfish_21092606_x64_avx2) 64bit 4CPU running on 2020 hardware recognizes the significance of Nxg5!! in 1 second. NNUE evaluation using nn-13406b1dcbe0.nnue enabled. 21/34 00:01 8291k 4530k +2,78 Nxg5 Bxd1 Nxe6 Qb8 Nxg7+ Kd8 Kxd1 b5 N3f5 Bf8 Rf1 Kc8 Nh5 Kb7 Bxb5 Ne7 g4 a6 Ba4 Nxf5 gxf5 Ka7 Nf4 c5 47/59 37:49 10390430k 4578k +3,16 Nxg5 Bxd1 Nxe6 Qb8 Nxg7+ Kd8 Kxd1 b5 Rf1 Kc8 N3f5 Bf8 Ne6 Kd7 Nf4 Ne7 g4 a5 Ke2 Qb7 h4 Ra6 a3 Kc8 Be3 Kb8 Kf3 Rb6 Bd2 Qc8 Kg3 c5 Be3 c4 Nxe7 Bxe7 Bf5 Qd8 h5 Qg8 Kh3 Bg5 Rf3 Ra6 Raf1 b4 Nxd5 Qxd5 Bxg5 bxc3 bxc3 Rb6 Be3 Rb3 Blacks 14 .. Kf8 is suboptimal and leads loss fast 41/68 06:31 3269727k 8350k +9,28 Bh6 Kg8 Rxd1 Bf8 N3h5 Bxg7 Nxg7 Qf8 Nf5 Ne7 Bxf8 Nxf5 Bxf5 Rxf8 Be6+ Kg7 Rd3 Rf4 Bxd5 c6 Rg3+ Kf8 Rf3 Rxf3 Bxf3 Kg7 Rf1 Re8 Be4 Re6 Ke2 a5 Ke3 Rh6 h3 a4 Kf4 Re6 h4 Re8 Ke3 h6 h5 Rf8 Rxf8 Kxf8 == Problem 4 == FEN: r1b1kb1r/1p1n1ppp/p2ppn2/6BB/2qNP3/2N5/PPP2PPP/R2Q1RK1 w kq - 0 1 10.Nxe6!! Qxe6 11.Nd5 Kd8 12.Bg4 Qe5 13.f4 Qxe4 (13...Qxb2 stronger but not sufficient: 14.Bxd7 Bxd7 15.Rb1 Qa3 16.Nxf6 Bb5 17.Qd4 Qc5 18.Rfd1 ±) 14.Bxd7 Bxd7 15.Nxf6 gxf6 16.Bxf6+ Kc7 17.Bxh8 and Black resigned on move 27. Stockfish 14dev 64bit 4CPU running on 2020 hardware recognises the significance of 10.Nxe6 in 1 second. Stockfish_21092606_x64_avx2: NNUE evaluation using nn-13406b1dcbe0.nnue enabled. 22/37 00:01 6955k 5367k +4,00 Nxe6 Qxe6 Nd5 Kd8 Bg4 Qe5 f4 Qxb2 Rb1 Qa3 Bxd7 Bxd7 Nxf6 Bb5 Rf3 Qxa2 c4 Bxc4 Rf2 Qa5 Nd5+ f6 Nxf6 Kc7 Rc1 b5 Qd5 gxf6 Bxf6 Kb8 Rxc4 Qe1+ Rf1 51/70 47:10 14538911k 5137k +5,76 Nxe6 Qxe6 Nd5 Kd8 Bg4 Qe5 f4 Qxe4 Bxd7 Bxd7 Nxf6 Qf5 Qd4 Kc8 Nd5 Bc6 c4 f6 Nb6+ Kb8 Bh4 Be7 Rae1 Bd8 Nxa8 Kxa8 Bf2 Kb8 Qxd6+ Bc7 Ba7+ Kc8 Qe6+ Qxe6 Rxe6 h5 h4 Rd8 Re7 g6 Be3 Ba5 Kf2 Rd6 Rc1 Bd8 Rg7 Be4 Rg8 Kd7 c5 Rd3 Rc4 Bd5 Rg7+ Ke6 Rd4 Rxd4 Bxd4 Kf5 Rd7 Bc6 Rxd8 Kxf4 Bxf6 == Problem 5 == FEN: r2qrb1k/1p1b2p1/p2ppn1p/8/3NP3/1BN5/PPP3QP/1K3RR1 w - - 0 1 21.e5!! dxe5 22.Ne4! Nh5 23.Qg6!? (stronger is 23.Qg4!! Nf4 24.Nf3 Qc7 25.Nh4 ± ) 23...exd4? (23...Nf4 24.Rxf4! exf4 25.Nf3! Qb6 26.Rg5!! covering b5 and threatening Nf6 or Ne5-f7+) 24.Ng5 1−0 Stockfish 8 64bit 3CPU running on 2016 hardware recognises the significance of 21.e5 in 5 seconds. Stockfish 12 dev (Stockfish_20062212_x64_modern) 64bit 1CPU running on 2016 hardware recognizes the significance of 21.e5 in 11 seconds. 25/42 00:06 7 963k 1309k +6,93 e5 Nh5 Ne4 dxe5 Nf3 Nf4 Qg4 Qc7 Nh4 Bc6 Nf6 g5 Rxf4 exf4 Qh5 Qe7 Ng6+ Kg7 Nxe7 Rxe7 Ng4 37/62 03:12 298 083k 1545k +10,70 e5 Ng4 Qxg4 Qg5 Qh3 Qxe5 Nde2 g5 Rxf8+ Kg7 Rff1 Rf8 Re1 Qf5 Qg3 Rad8 Nd4 Qf4 Nxe6+ Bxe6 Rxe6 Qxg3 == Problem 6 == FEN: rnbqk2r/1p3ppp/p7/1NpPp3/QPP1P1n1/P4N2/4KbPP/R1B2B1R b kq - 0 1 13... axb5!! offers an exchange to keep the white queen out of play. 14.Qxa8 Bd4 15.Nxd4 cxd4 16.Qxb8 0-0! 17.Ke1 Qh4 18.g3 Qf6 19.Bf4 g5? (Ivanchuk found 19...d3! during post-game analysis.) 20.Rc1 exf4 21.Qxf4 Qd4 22.Rd1 bxc4 23.e5 Qc3+ 24.Rd2 Re8 25.Bxd3 cxd3 −+ Tasc R30 finds 19... d3! in 2 1/2 hours. 19... Bf5!! is even stronger than 19... d3. Position is already lost at 19... d3 +8.00 for black, ... Bf5 not much better Stockfish 14dev 64bit 4CPU running on 2020 hardware recognises the significance of axb5!! in 1 second. Stockfish_21092606_x64_avx2: NNUE evaluation using nn-13406b1dcbe0.nnue enabled. 21/28 00:01 9264k 4714k -1,22 axb5 Qxa8 Bd4 Nxd4 cxd4 h3 Nf6 Bg5 0-0 cxb5 h6 Bxf6 Qxf6 Re1 Nd7 Kd1 Qg6 Qa4 Qg3 Qc2 Qxa3 Bd3 Qxb4 Qb1 46/67 1:05:00 18113493k 4644k -2,40 axb5 Qxa8 Bd4 h3 Nf6 Nxd4 exd4 Kf2 Nxe4+ Kg1 Nd7 Bg5 Qxg5 Qxc8+ Ke7 Qc7 Qe5 d6+ Qxd6 Qxd6+ Kxd6 bxc5+ Ndxc5 cxb5 d3 h4 d2 Rh3 Ke5 Be2 f5 Ra2 Rd8 Bd1 Rd4 Re3 f4 Re2 b6 a4 Kd6 Rc2 Kd5 Ra2 h6 Rb2 Nxa4 Bxa4 Rxa4 Rexd2+ Nxd2 Rxd2+ Kc4 Rd7 g6 == Problem 7 == FEN 1r1bk2r/2R2ppp/p3p3/1b2P2q/4QP2/4N3/1B4PP/3R2K1 w k - 0 1 1.Rxd8+!! Rxd8 (1...Kxd8 2.Ra7! Qe2 3.Qd4+ Ke8 4.h3 Qe1+ 5.Kh2 Rd8 6.Qc5 Qh4 7.Ba3 Rd7 8.Ra8+ Rd8 9.g3 1−0)

  Read more →
• 

  Hugging Face

  Hugging Face, Inc., is an American company based in New York City that develops computation tools for building applications using machine learning. Its transformers library built for natural language processing applications and its platform allow users to share machine learning models and datasets and showcase their work. == History == === Founding === The company was founded in 2016 by French entrepreneurs Clément Delangue, Julien Chaumond, and Thomas Wolf in New York City, originally as a company that developed a chatbot app targeted at teenagers. The company was named after the U+1F917 🤗 HUGGING FACE emoji. After open sourcing the model behind the chatbot, the company pivoted to focus on being a platform for machine learning. === AI boom === On April 28, 2021, the company launched the BigScience Research Workshop in collaboration with several other research groups to release an open large language model. In 2022, the workshop concluded with the announcement of BLOOM, a multilingual large language model with 176 billion parameters. In February 2023, the company announced partnership with Amazon Web Services (AWS) which would allow Hugging Face's products to be available to AWS customers to use them as the building blocks for their custom applications. The company also said the next generation of BLOOM will be run on Trainium, a proprietary machine learning chip created by AWS. In June 2024, the company announced, along with Meta and Scaleway, their launch of a new AI accelerator program for European startups. The initiative aimed to help startups integrate open foundation models into their products, accelerating the EU AI ecosystem. The program, based at STATION F in Paris, ran from September 2024 to February 2025. Selected startups received mentoring, and access to AI models and tools and Scaleway's computing power. On September 23, 2024, to further the International Decade of Indigenous Languages, Hugging Face teamed up with Meta and UNESCO to launch a new online language translator. It was built on Meta's No Language Left Behind open-source AI model, enabling free text translation across 200 languages, including many low-resource languages. In April 2025, Hugging Face announced that they acquired a humanoid robotics startup, Pollen Robotics, based in France and founded by Matthieu Lapeyre and Pierre Rouanet in 2016. In an X tweet, Delangue shared his vision to "make Artificial Intelligence robotics Open Source". === Cyberattacks === In early 2026, hackers hijacked the Hugging Face platform to launch Android-targeted attacks involving "powerful malware" which could completely take over a compromised target.

  Read more →
• 

  AI data center

  An AI data center is a specialized data center facility designed for the computationally intensive tasks of training and running inference for artificial intelligence (AI) and machine learning models. Unlike general-purpose data centers, they are optimized for the parallel processing demands of AI workloads, typically using hardware such as AI accelerators (e.g., GPUs, TPUs) and high-speed interconnects. The global push to construct these specialized facilities accelerated dramatically during the AI boom of the 2020s. Memory manufacturers prioritized production of High Bandwidth Memory (HBM) essential for AI servers, which led to a global memory supply shortage amid a broader competition for advanced chips, power, and infrastructure. Major tech companies are estimated to spend $650 billion on AI data centers in 2026. == Architecture == Data centers for building and running large machine learning models contain specialized computer chips, GPUs, that use 2 to 4 times as much energy as their regular CPU counterparts (250-500 watts). AI data centers use 60 or more kilowatts per server rack, whereas more standard data centers typically use 5 to 10 kilowatts per rack. == Operators == As of August 2025, The Information tracked 18 planned or existing AI data centers in the United States, operated by Amazon Web Services, CoreWeave, Crusoe, Meta, Microsoft/OpenAI, Oracle, Tesla, and xAI. Other AI data center operators include Digital Realty and Alibaba. Data centers are also being built in China, India, Europe, Saudi Arabia, and Canada. The New Yorker described CoreWeave as the most prominent AI data center operator in the United States. Two types of data center providers for machine learning have been noted: hyperscalers and neoclouds. The Verge listed large technology companies such as Google, Meta, Microsoft, Oracle and Amazon as hyperscalers. The New York Times described neoclouds as "a new generation of data center providers". CoreWeave, Nebius, Nscale, and Lambda have been described as examples of neoclouds. In January 2025, OpenAI, in partnership with Oracle and Softbank, announced the Stargate project, which as of September 2025 is composed of six built or proposed AI data centers in the United States. In response to the Stargate project, Amazon launched in October 2025 an AI data center on 1,200 acres of farmland in Indiana. This data center, known as Project Rainier, is one of the largest AI data centers in the world, with Amazon spending $11 billion on the project. Rainier is specifically intended for training and running machine learning models from Anthropic. As of that time, this facility contains seven data centers (out of an estimated 30 planned) and will use 2.2 gigawatts of electricity (equivalent to 1 million households) and millions of gallons of water per year. Computer chips from Annapurna Labs and Anthropic, Trainium 2, were designed for use in such facilities. Amazon pumped millions of gallons of water out of the ground to construct the data center, and as of June 2025, Indiana state officials are investigating whether this dewatering process led to dry wells for local residents. In November 2025, Anthropic announced a plan in partnership with Fluidstack to develop artificial intelligence infrastructure in the United States, including data centers in New York and Texas, worth $50 billion. Other AI data center projects include the Colossus supercomputer from xAI, a Louisiana-based project from Meta, Hyperion, expected to use 5 GW of power, and a second Ohio-based Meta project, Prometheus, with a capacity of 1 GW. A 3,200-acre AI data center, capable of 4.4-4.5 GW of power and located on the decommissioned Homer City Generating Station, is under construction as of 2025, and will use seven 30-acre gas generating stations supplied by EQT. As of December 2025, CRH is working on over 100 data centers in the United States. In 2025, ExxonMobil and NextEra announced plans to build a data center powered by natural gas and using carbon capture technology, with 1.2 GW of power capacity. They previously purchased 2,500 acres of land in the Southeastern United States and plan to market the data center to an artificial intelligence company. The increased interest in AI data centers has led to several executives from companies in that space becoming billionaires, including CoreWeave, QTS, Nebius, Astera Labs, Groq, Fermi (which is connected to former United States Secretary of Energy Rick Perry), Snowflake and Cipher Mining. Several companies involved in cryptocurrency mining, such as Bitdeer, CoreWeave, Cipher Mining, TeraWulf, IREN, Core Scientific, and CleanSpark have also been involved with AI data centers. == Finances == Between January and August 2024, Microsoft, Meta, Google and Amazon collectively spent $125 billion on AI data centers. Citigroup forecasted that $2.8 trillion would be spent on AI data centers by 2030, while McKinsey and Company estimated that almost $7 trillion would be spent globally by that time. According to S&P Global, $61 billion has been spent on the data center market as a whole in 2025, while debt issuance for data centers was $182 billion during the same year. Large technology companies have offloaded the financial risks of building AI data centers by setting up special purpose vehicles or by contracting with neoclouds. For example, Meta's Hyperion was mostly funded by Blue Owl Capital, which did so using a bond offering from PIMCO. Those bonds were sold to a number of clients, including BlackRock. Meta did not borrow money itself and instead established a special purpose vehicle from which it would rent the data center. This deal was structured by Morgan Stanley for $30 billion, the largest known private capital transaction as of 2025. Neoclouds such as CoreWeave have gone into debt to buy computer chips from Nvidia for their data centers, and the chips themselves have been used for loan collateral. As of December 2025, CoreWeave took out three GPU-backed loans, collectively worth $12.4 billion, from private credit firms (Blackstone, Coatue, BlackRock, PIMCO) and from banks (Goldman Sachs, JPMorgan Chase, Wells Fargo). Thus, these companies provide an indirect connection between private credit and established banks. Data centers have also established asset-backed securities, and debt for data centers has its own derivative financial products. The real estate industry, including asset managers, public companies and private investors, has also invested in data centers. == Energy sourcing == == Environmental footprint == Average AI data centers have an electricity footprint equivalent to 100,000 households, and use billions of gallons of water for cooling their hardware. In 2025, the International Energy Agency estimated that the larger AI data centers currently under construction could consume as much electricity as 2 million households. A 2024 report from the United States Department of Energy stated that data centers overall used 17 billion gallons of water per year in the United States, primarily due to "rapid proliferation of AI servers", and that this usage was forecasted to grow to nearly 80 billion gallons by 2028. Researchers estimated that AI data centers in the United States would emit 24-44 million metric tons of carbon dioxide and use 731–1,125 million cubic meters of water per year between 2024 and 2030. Peaking power plants, which have been proposed as a power source for AI data centers, emit sulfur dioxide and have historically been located disproportionately near communities of color in the United States. Reciprocating internal combustion engines, proposed as another power source for a data center, emit PM 2.5, nitrogen oxides, and volatile organic compounds. == AI data centers in the United States == In the United States, both the Biden administration and second Trump administration supported the construction of AI data centers. In January 2025, then-president Joe Biden signed an executive order for federal government agencies to support AI data centers on federal sites built by private companies, study their effect on energy prices, and encourage their use of renewable energy. In April 2025, the United States Department of Energy suggested 16 possible sites, including Los Alamos National Laboratory, Sandia National Laboratories and Oak Ridge National Laboratory. In its July 2025 AI Action Plan, the second Trump administration supported increased production of AI data centers. Several US states have incentivized local data center construction. For example, in 2024, lawmakers in Michigan approved tax breaks for data center equipment and construction material. Some data center companies have also invested or promised to invest in the infrastructure of local communities. In December 2025, Democratic senators Elizabeth Warren, Chris Van Hollen, and Richard Blumenthal wrote to seven technology companies (Google, Microsoft, Amazon, Meta, CoreWeave, Digital Realty, and Equinix) that they w

  Read more →
• 

  Meesho

  Meesho Limited (short for Meri shop, transl. My shop) is an Indian e-commerce company, headquartered in Bengaluru. Founded by Vidit Aatrey and Sanjeev Barnwal in December 2015, Meesho is an online marketplace in categories such as fashion, home and kitchen, beauty and personal care, electronics accessories, and daily use products. == History == Meesho Private Limited, formerly Fashnear Technologies Private Limited, was established by IIT Delhi graduates Vidit Aatrey and Sanjeev Barnwal in December, 2015 In 2016, the founders came up with the idea of re-establishing the platform as Meesho, one that would enable country-wide shipping for resellers with the use of social media sites as tools for marketing. In February 2019, the platform reported having around 209,000 users and about 1.2 million monthly orders, and in March 2020, it reported approximately 563,000 users and 3.1 million monthly orders. In 2021, the Meesho mobile application was ranked among the most downloaded shopping apps globally. In 2022, Meesho had about 120 million monthly users and about 910 million orders were made through the platform, with a gross merchandise value (GMV) of about $5 billion. According to report as of August 2023 Meesho delisted 42 lakh counterfeit listings and 10 lakh restricted products under its initiative Project Suraksha. During the same period, the platform blocked access for over 12,000 user accounts flagged for policy violations. The Court granted injunctive relief by directing domain registrars to suspend the infringing websites. Additionally, the Court ordered law enforcement authorities to initiate criminal investigations, freeze associated financial accounts against the identified offenders. In 2023, Meesho became the fastest shopping app to cross over 500 million downloads. In 2024, Meesho introduced Valmo, a logistics marketplace, to provide shipment services to sellers by aggregating multiple logistics providers. Meesho employs over 3,000 small businesses and 10-12 large firms for warehousing and sorting operations within its logistics framework. According to media reports, Valmo operating in approximately 15,000 pincodes in India with around 6,000 partners. It is reported to handle over 50% of Meesho's daily orders. In November 2024, Meesho introduced a generative AI-powered voice bot for customer support, managing approximately 60,000 calls daily in English and Hindi. According to media reports, the system resolves the majority of queries without human assistance, with only a small fraction of calls requiring manual intervention. According to media reports, in 2024, Meesho prevented over 22 million suspicious or potentially fraudulent transactions on its platform. The company initiated legal proceedings, resulting in the filing of twelve cases, including nine specifically targeting over forty individuals in the cities of Kolkata and Ranchi. The company filed a suit in the Delhi High Court for a permanent injunction against parties operating deceptive websites misappropriating its brand identity. Meesha went public through an initial public offering in December 2025, raising $603 million. It is listed on both the BSE and NSE. == Recognition == In 2023, Meesho was named one of the most influential companies of the year by Time (magazine).

  Read more →
• 

  INDIAai

  INDIAai is a web portal launched by the Government of India on 07 March 2024 for artificial intelligence-related developments in India. It is known as the National AI Portal of India, which was jointly started by the Ministry of Electronics and Information Technology (MeitY), the National e-Governance Division (NeGD) and the National Association of Software and Service Companies (NASSCOM) with support from the Department of School Education and Literacy (DoSE&L) and Ministry of Human Resource Development. == History == The portal was launched on 30 May 2020, by Ravi Shankar Prasad, the Union Minister for Electronics and IT, Law and Justice and Communications, on the first anniversary of the second tenure of Prime Minister Narendra Modi-led government. A national program for the youth, 'Responsible AI for Youth', was also launched on the same day. As of 2022, the website was visited by more than 4.5 lakh users with 1.2 million page views. It has 1151 articles on artificial intelligence, 701 news stories, 98 reports, 95 case studies and 213 videos on its portal. It maintains a database on AI ecosystem of India featuring 121 government initiatives and 281 startups. In May 2022, INDIAai released a book titled 'AI for Everyone' that covers the basics of AI. Cabinet chaired by the Prime Minister Narendra Modi has approved the comprehensive national-level IndiaAI mission with a budget outlay of Rs.10,371.92 crore. The Mission will be implemented by ‘IndiaAI’ Independent Business Division (IBD) under Digital India Corporation (DIC). == Objective and features == It aims to function as a one-stop portal for all AI-related development in India. The platform publishes resources such as articles, news, interviews, and investment funding news and events for AI startups, AI companies, and educational firms related to artificial intelligence in India. It also distributes documents, case studies, and research reports. Additionally, the platform provides education and employment opportunities related to AI. It offers AI courses, both free and paid.

  Read more →
• 

  Empowerment (artificial intelligence)

  Empowerment in the field of artificial intelligence formalises and quantifies (via information theory) the potential an agent perceives that it has to influence its environment. An agent which follows an empowerment maximising policy, acts to maximise future options (typically up to some limited horizon). Empowerment can be used as a (pseudo) utility function that depends only on information gathered from the local environment to guide action, rather than seeking an externally imposed goal, thus is a form of intrinsic motivation. The empowerment formalism depends on a probabilistic model commonly used in artificial intelligence. An autonomous agent operates in the world by taking in sensory information and acting to change its state, or that of the environment, in a cycle of perceiving and acting known as the perception-action loop. Agent state and actions are modelled by random variables ( S : s ∈ S , A : a ∈ A {\displaystyle S:s\in {\mathcal {S}},A:a\in {\mathcal {A}}} ) and time ( t {\displaystyle t} ). The choice of action depends on the current state, and the future state depends on the choice of action, thus the perception-action loop unrolled in time forms a causal bayesian network. == Definition == Empowerment ( E {\displaystyle {\mathfrak {E}}} ) is defined as the channel capacity ( C {\displaystyle C} ) of the actuation channel of the agent, and is formalised as the maximal possible information flow between the actions of the agent and the effect of those actions some time later. Empowerment can be thought of as the future potential of the agent to affect its environment, as measured by its sensors. E := C ( A t ⟶ S t + 1 ) ≡ max p ( a t ) I ( A t ; S t + 1 ) {\displaystyle {\mathfrak {E}}:=C(A_{t}\longrightarrow S_{t+1})\equiv \max _{p(a_{t})}I(A_{t};S_{t+1})} In a discrete time model, Empowerment can be computed for a given number of cycles into the future, which is referred to in the literature as 'n-step' empowerment. E ( A t n ⟶ S t + n ) = max p ( a t , . . . , a t + n − 1 ) I ( A t , . . . , A t + n − 1 ; S t + n ) {\displaystyle {\mathfrak {E}}(A_{t}^{n}\longrightarrow S_{t+n})=\max _{p(a_{t},...,a_{t+n-1})}I(A_{t},...,A_{t+n-1};S_{t+n})} The unit of empowerment depends on the logarithm base. Base 2 is commonly used in which case the unit is bits. === Contextual Empowerment === In general the choice of action (action distribution) that maximises empowerment varies from state to state. Knowing the empowerment of an agent in a specific state is useful, for example to construct an empowerment maximising policy. State-specific empowerment can be found using the more general formalism for 'contextual empowerment'. C {\displaystyle C} is a random variable describing the context (e.g. state). E ( A t n ⟶ S t + n ∣ C ) = ∑ c ∈ C p ( c ) E ( A t n ⟶ S t + n ∣ C = c ) {\displaystyle {\mathfrak {E}}(A_{t}^{n}\longrightarrow S_{t+n}{\mid }C)=\sum _{c{\in }C}p(c){\mathfrak {E}}(A_{t}^{n}\longrightarrow S_{t+n}{\mid }C=c)} == Application == Empowerment maximisation can be used as a pseudo-utility function to enable agents to exhibit intelligent behaviour without requiring the definition of external goals, for example balancing a pole in a cart-pole balancing scenario where no indication of the task is provided to the agent. Empowerment has been applied in studies of collective behaviour and in continuous domains. As is the case with Bayesian methods in general, computation of empowerment becomes computationally expensive as the number of actions and time horizon extends, but approaches to improve efficiency have led to usage in real-time control. Empowerment has been used for intrinsically motivated reinforcement learning agents playing video games, and in the control of underwater vehicles.

  Read more →
• 

  Automated negotiation

  Automated negotiation is a form of interaction in systems that are composed of multiple autonomous agents, in which the aim is to reach agreements through an iterative process of making offers. Automated negotiation can be employed for many tasks human negotiators regularly engage in, such as bargaining and joint decision making. The main topics in automated negotiation revolve around the design of protocols and negotiating strategies. == History == Through digitization, the beginning of the 21st century has seen a growing interest in the automation of negotiation and e-negotiation systems, for example in the setting of e-commerce. This interest is fueled by the promise of automated agents being able to negotiate on behalf of human negotiators, and to find better outcomes than human negotiators. == Examples == Examples of automated negotiation include: Online dispute resolution, in which disagreements between parties are settled. Sponsored search auction, where bids are placed on advertisement keywords. Content negotiation, in which user agents negotiate over HTTP about how to best represent a web resource. Negotiation support systems, in which negotiation decision-making activities are supported by an information system.

  Read more →
• 

  Straight-Through Quality

  Straight-Through Quality (STQ) are approaches and outputs of test automation that have quality and deliver business benefit. STQ takes its name from the business concept of straight-through processing (STP). Also acting as a tool and enabler for STP. Traditional techniques for testing and delivery have often required a great deal of manual support and intervention. These approaches are subject to human error, cost of delay and lack of reuse. These also have the negative side-effect of being unable to deliver 'fail-fast' approaches, which have proven popular with Agile practitioners. Previous traditional approaches have been typically expensive where whole silo'ed departments are created within commercial companies to deliver Quality and Deployment alone. Thus STQ as an approach hopes to resolve this problem. == Examples == Tangible examples of STQ approaches in the software industry are present and often known as continuous integration (CI) and continuous delivery (CD). These combined can ensure that software delivery is integrated, automatically tested and ready for automatic delivery at any time. Together CI/CD can enable STQ which can be used as Business output terminology for business users who do not understand the technical complexities of CI/CD.

  Read more →
• 

  Hierarchical Risk Parity

  Hierarchical Risk Parity (HRP) is an advanced investment portfolio optimization framework developed in 2016 by Marcos López de Prado at Guggenheim Partners and Cornell University. HRP is a probabilistic graph-based alternative to the prevailing mean-variance optimization (MVO) framework developed by Harry Markowitz in 1952, and for which he received the Nobel Prize in economic sciences. HRP algorithms apply discrete mathematics and machine learning techniques to create diversified and robust investment portfolios that outperform MVO methods out-of-sample. HRP aims to address the limitations of traditional portfolio construction methods, particularly when dealing with highly correlated assets. Following its publication, HRP has been implemented in numerous open-source libraries, and received multiple extensions. == Key features == HRP portfolios have been proposed as a robust alternative to traditional quadratic optimization methods, including the Critical Line Algorithm (CLA) of Markowitz. HRP addresses three central issues commonly associated with quadratic optimizers: numerical instability, excessive concentration in a small number of assets, and poor out-of-sample performance. HRP leverages techniques from graph theory and machine learning to construct diversified portfolios using only the information embedded in the covariance matrix. Unlike quadratic programming methods, HRP does not require the covariance matrix to be invertible. Consequently, HRP remains applicable even in cases where the covariance matrix is ill-conditioned or singular—conditions under which standard optimizers fail. Monte Carlo simulations indicate that HRP achieves lower out-of-sample variance than CLA, despite the fact that minimizing variance is the explicit optimization objective of CLA. Furthermore, HRP portfolios exhibit lower realized risk compared to those generated by traditional risk parity methodologies. Empirical backtests have demonstrated that HRP would have historically outperformed conventional portfolio construction techniques. Algorithms within the HRP framework are characterized by the following features: Machine Learning Approach: HRP employs hierarchical clustering, a machine learning technique, to group similar assets based on their correlations. This allows the algorithm to identify the underlying hierarchical structure of the portfolio, and avoid that errors spread through the entire network. Risk-Based Allocation: The algorithm allocates capital based on risk, ensuring that assets only compete with similar assets for representation in the portfolio. This approach leads to better diversification across different risk sources, while avoiding the instability associated with noisy returns estimates. Covariance Matrix Handling: Unlike traditional methods like Mean-Variance Optimization, HRP does not require inverting the covariance matrix. This makes it more stable and applicable to portfolios with a large number of assets, particularly when the covariance matrix's condition number is high. == The problem: Markowitz's Curse == Portfolio construction is perhaps the most recurrent financial problem. On a daily basis, investment managers must build portfolios that incorporate their views and forecasts on risks and returns. Despite the theoretical elegance of Markowitz's mean-variance framework, its practical implementation is hindered by several limitations that undermine the reliability of solutions derived from the Critical Line Algorithm (CLA). A principal concern is the high sensitivity of optimal portfolios to small perturbations in expected returns: even minor forecasting errors can result in significantly different allocations (Michaud, 1998). Given the inherent difficulty of producing accurate return forecasts, numerous researchers have advocated for approaches that forgo expected returns entirely and instead rely solely on the covariance structure of asset returns. This has given rise to risk-based allocation methods, among which risk parity is a widely cited example (Jurczenko, 2015). While eliminating return forecasts mitigates some instability, it does not eliminate it. Quadratic programming techniques employed in portfolio optimization require the inversion of a positive-definite covariance matrix, meaning all eigenvalues must be strictly positive. When the matrix is numerically ill-conditioned—that is, when the ratio of its largest to smallest eigenvalue (its condition number) is large—matrix inversion becomes unreliable and prone to significant numerical errors (Bailey and López de Prado, 2012). The condition number of a covariance, correlation, or any symmetric (and thus diagonalizable) matrix is defined as the absolute value of the ratio between its largest and smallest eigenvalues in modulus. The figure on the right presents the sorted eigenvalues of several correlation matrices; the condition number is represented by the ratio of the first to last eigenvalues in each sequence. A diagonal correlation matrix, which is equal to its own inverse, exhibits the minimum possible condition number. As the number of correlated (or multicollinear) assets in a portfolio increases, the condition number rises. At high levels, this leads to severe numerical instability, whereby slight modifications in any matrix entry may result in drastically different inverses. This phenomenon, often referred to as Markowitz’s curse, encapsulates the paradox wherein increased correlation among assets heightens the theoretical need for diversification, yet simultaneously increases the likelihood of unstable optimization outcomes. Consequently, the potential benefits of diversification are frequently overshadowed by estimation errors. These problems are exacerbated as the dimensionality of the covariance matrix increases. The estimation of each covariance term consumes degrees of freedom, and in general, a minimum of 1 2 N ( N + 1 ) {\displaystyle {\frac {1}{2}}N(N+1)} independent and identically distributed (IID) observations is required to estimate a non-singular covariance matrix of dimension N {\displaystyle N} . For example, constructing an invertible covariance matrix of dimension 50 necessitates at least five years of daily IID observations. However, empirical evidence suggests that the correlation structure of financial assets is highly unstable over such extended periods. These difficulties are highlighted by the observation that even naïve allocation strategies—such as equally weighted portfolios—have frequently outperformed both mean-variance and risk-based optimizations in out-of-sample tests (De Miguel et al., 2009). == The solution: Hierarchical Risk Parity == The HRP algorithm addresses Markowitz's curse in three steps: Hierarchical Clustering: Assets are grouped into clusters based on their correlations, forming a hierarchical tree structure. Quasi-Diagonalization: The correlation matrix is reordered based on the clustering results, revealing a block diagonal structure. Recursive Bisection: Weights are assigned to assets through a top-down approach, splitting the portfolio into smaller sub-portfolios and allocating capital based on inverse variance. === Step 1: Hierarchical clustering === Given a T × N {\displaystyle T\times N} matrix of asset returns X {\displaystyle X} , where each column represents a time series of returns for one of N {\displaystyle N} assets over T {\displaystyle T} time periods, a hierarchical clustering process can be used to construct a tree-based representation of asset relationships. First, we compute the N × N {\displaystyle N\times N} correlation matrix ρ = ρ i , j i , j = 1 . . . N {\displaystyle \rho ={\rho _{i,j}}\;{i,j=1\;...\;N}} , where ρ i , j = c o r r ( X i , X j ) {\displaystyle \rho _{i,j}=\mathrm {corr} (X_{i},X_{j})} . From this, a pairwise distance matrix D = d i , j {\displaystyle D={d_{i,j}}} is defined using the transformation: d i , j = 1 2 ( 1 − ρ i , j ) {\displaystyle d_{i,j}={\sqrt {{\frac {1}{2}}(1-\rho _{i,j})}}} This distance function defines a proper metric space, satisfying non-negativity, identity of indiscernibles, symmetry, and the triangle inequality. Next, a secondary distance matrix D ~ = d ~ i , j {\displaystyle {\tilde {D}}={{\tilde {d}}_{i,j}}} is computed, where each entry measures the Euclidean distance between the distance profiles of two assets: d ~ i , j = ∑ n = 1 N ( d n , i − d n , j ) 2 {\displaystyle {\tilde {d}}_{i,j}={\sqrt {\sum _{n=1}^{N}(d_{n,i}-d_{n,j})^{2}}}} While d i , j {\displaystyle d_{i,j}} reflects correlation-based proximity between two assets, d ~ i , j {\displaystyle {\tilde {d}}_{i,j}} quantifies dissimilarity across the entire system, as it depends on all pairwise distances. Hierarchical clustering proceeds by identifying the pair ( i , j ) {\displaystyle (i,j)} with the smallest value of d ~ i , j {\displaystyle {\tilde {d}}_{i,j}} (for i ≠ j {\displaystyle i\neq j} ), and forming a new cluster u [ 1 ] = ( i , j ) {\displaystyle u[1]=(i,j)} .

  Read more →
• 

  AI Security Institute

  The AI Security Institute (AISI) is a research organisation under the Department for Science, Innovation and Technology, UK, that aims "to equip governments with a scientific understanding of the risks posed by advanced AI". It conducts research and develop and test mitigations. Previously, it was known as the AI Safety Institute. Its creation followed world's first major AI Safety Summit that was held in Bletchley Park in 2023. The institute's professed goal is "building the world's leading understanding of advanced AI risks and solutions, to inform governments so they can keep the public safe". It is designed like a startup in the government "combining the authority of government with the expertise and agility of the private sector". AISI has made access agreements with Anthropic, Google and OpenAI to test their models before release. It has an open source platform called Inspect that permits companies, governments and academics to run standardised safety tests for AI usage. Among the works AISI has done is the reported detection of multiple serious vulnerabilities that could enable development of biological weapons; the vulnerabilities were fixed before the model was launched. It conducts research on diverse fields of AI application. One study by AISI found that LLMs post-trained for political persuasiveness became systematically less accurate and up to 51% more persuasive on political issues. AISI has also worked on the usage of AI for emotional needs. It found that nearly 10 percent of UK citizens used systems like chatbots for emotional purposes on a weekly basis. It found that "systems are now outperforming PhD-level researchers on scientific knowledge tests and helping non-experts succeed at lab work that would previously have been out of reach" in a report published in December 2025. Former chief AI officer of GCHQ Adam Beaumont is the institution's interim director. UK prime minister's AI advisor Jade Leung is the chief technology officer.

  Read more →