AI Face Combiner

AI Face Combiner — independent reviews, comparisons, pricing and step-by-step guides on Aizhi.

• 

  Non-local means

  Non-local means is an algorithm in image processing for image denoising. Unlike "local mean" filters, which take the mean value of a group of pixels surrounding a target pixel to smooth the image, non-local means filtering takes a mean of all pixels in the image, weighted by how similar these pixels are to the target pixel. This results in much greater post-filtering clarity, and less loss of detail in the image compared with local mean algorithms. If compared with other well-known denoising techniques, non-local means adds "method noise" (i.e. error in the denoising process) which looks more like white noise, which is desirable because it is typically less disturbing in the denoised product. Recently non-local means has been extended to other image processing applications such as deinterlacing, view interpolation, and depth maps regularization. == Definition == Suppose Ω {\displaystyle \Omega } is the area of an image, and p {\displaystyle p} and q {\displaystyle q} are two points within the image. Then, the algorithm is: u ( p ) = 1 C ( p ) ∫ Ω v ( q ) f ( p , q ) d q . {\displaystyle u(p)={1 \over C(p)}\int _{\Omega }v(q)f(p,q)\,\mathrm {d} q.} where u ( p ) {\displaystyle u(p)} is the filtered value of the image at point p {\displaystyle p} , v ( q ) {\displaystyle v(q)} is the unfiltered value of the image at point q {\displaystyle q} , f ( p , q ) {\displaystyle f(p,q)} is the weighting function, and the integral is evaluated ∀ q ∈ Ω {\displaystyle \forall q\in \Omega } . C ( p ) {\displaystyle C(p)} is a normalizing factor, given by C ( p ) = ∫ Ω f ( p , q ) d q . {\displaystyle C(p)=\int _{\Omega }f(p,q)\,\mathrm {d} q.} == Common weighting functions == The purpose of the weighting function, f ( p , q ) {\displaystyle f(p,q)} , is to determine how closely related the image at the point p {\displaystyle p} is to the image at the point q {\displaystyle q} . It can take many forms. === Gaussian === The Gaussian weighting function sets up a normal distribution with a mean, μ = B ( p ) {\displaystyle \mu =B(p)} and a variable standard deviation: f ( p , q ) = e − | B ( q ) − B ( p ) | 2 h 2 {\displaystyle f(p,q)=e^{-{{\left\vert B(q)-B(p)\right\vert ^{2}} \over h^{2}}}} where h {\displaystyle h} is the filtering parameter (i.e., standard deviation) and B ( p ) {\displaystyle B(p)} is the local mean value of the image point values surrounding p {\displaystyle p} . == Discrete algorithm == For an image, Ω {\displaystyle \Omega } , with discrete pixels, a discrete algorithm is required. u ( p ) = 1 C ( p ) ∑ q ∈ Ω v ( q ) f ( p , q ) {\displaystyle u(p)={1 \over C(p)}\sum _{q\in \Omega }v(q)f(p,q)} where, once again, v ( q ) {\displaystyle v(q)} is the unfiltered value of the image at point q {\displaystyle q} . C ( p ) {\displaystyle C(p)} is given by: C ( p ) = ∑ q ∈ Ω f ( p , q ) {\displaystyle C(p)=\sum _{q\in \Omega }f(p,q)} Then, for a Gaussian weighting function, f ( p , q ) = e − | B ( q ) 2 − B ( p ) 2 | h 2 {\displaystyle f(p,q)=e^{-{{\left\vert B(q)^{2}-B(p)^{2}\right\vert } \over h^{2}}}} where B ( p ) {\displaystyle B(p)} is given by: B ( p ) = 1 | R ( p ) | ∑ i ∈ R ( p ) v ( i ) {\displaystyle B(p)={1 \over |R(p)|}\sum _{i\in R(p)}v(i)} where R ( p ) ⊆ Ω {\displaystyle R(p)\subseteq \Omega } and is a square region of pixels surrounding p {\displaystyle p} and | R ( p ) | {\displaystyle |R(p)|} is the number of pixels in the region R {\displaystyle R} . == Efficient implementation == The computational complexity of the non-local means algorithm is quadratic in the number of pixels in the image, making it particularly expensive to apply directly. Several techniques were proposed to speed up execution. One simple variant consists of restricting the computation of the mean for each pixel to a search window centred on the pixel itself, instead of the whole image. Another approximation uses summed-area tables and fast Fourier transform to calculate the similarity window between two pixels, speeding up the algorithm by a factor of 50 while preserving comparable quality of the result.

  Read more →
• 

  Parallel terraced scan

  The parallel terraced scan is a multi-agent based search technique that is basic to cognitive architectures, such as Copycat, Letter-string, the Examiner, Tabletop, and others. It was developed by John Rehling and Douglas Hofstadter at the Center for Research on Concepts and Cognition at Indiana University, Bloomington. The parallel terraced scan builds on the concepts of the workspace, coderack, conceptual memory, and temperature. According to Hofstadter the parallel and random nature of the processing captures aspects of human cognition.

  Read more →
• 

  Preferential entailment

  Preferential entailment is a non-monotonic logic based on selecting only models that are considered the most plausible. The plausibility of models is expressed by an ordering among models called a preference relation, hence the name preference entailment. Formally, given a propositional formula F {\displaystyle F} and an ordering over propositional models ≤ {\displaystyle \leq } , preferential entailment selects only the models of F {\displaystyle F} that are minimal according to ≤ {\displaystyle \leq } . This selection leads to a non-monotonic inference relation: F ⊨ pref G {\displaystyle F\models _{\text{pref}}G} holds if and only if all minimal models of F {\displaystyle F} according to ≤ {\displaystyle \leq } are also models of G {\displaystyle G} . Circumscription can be seen as the particular case of preferential entailment when the ordering is based on containment of the sets of variables assigned to true (in the propositional case) or containment of the extensions of predicates (in the first-order logic case).

  Read more →
• 

  Utah Artificial Intelligence Policy Act

  The Utah Artificial Intelligence Policy Act (SB-149) was signed into law in Utah in 2024 and amended in 2025. The first state law in the United States specifically regulating generative AI, it went into effect on May 1, 2024. The law requires companies to disclose if their customers interact with AI instead of a human. It also established an Office of Artificial Intelligence Policy. Amendments to the Act went into effect on May 7, 2025. While the 2024 Act requires companies to disclose generative AI use when asked by customers, the amendments introduced stricter requirements for higher-risk interactions. SB 226 mandates disclosure of AI use in high-risk interactions involving health, financial, and biometric data, or when providing consumers with advice on financial, legal, or healthcare matters.

  Read more →
• 

  Embedding (machine learning)

  In machine learning, embedding is a representation learning technique that maps complex, high-dimensional data into a lower-dimensional vector space of numerical vectors. == Technique == It also denotes the resulting representation, where meaningful patterns or relationships are preserved. As a technique, it learns these vectors from data like words, images, or user interactions, differing from manually designed methods such as one-hot encoding. This process reduces complexity and captures key features without needing prior knowledge of the domain. == Similarity == In natural language processing, words or concepts may be represented as feature vectors, where similar concepts are mapped to nearby vectors. The resulting embeddings vary by type, including word embeddings for text (e.g., Word2Vec), image embeddings for visual data, and knowledge graph embeddings for knowledge graphs, each tailored to tasks like NLP, computer vision, or recommendation systems. This dual role enhances model efficiency and accuracy by automating feature extraction and revealing latent similarities across diverse applications. To measure the distance between two embeddings, a similarity measure can be used to find the overall similarity of the concepts represented by the embeddings. If the vectors are normalized to have a magnitude of 1, then the similarity measures are proportional to cos ⁡ ( θ a b ) {\displaystyle \cos \left(\theta _{ab}\right)} . The cosine similarity disregards the magnitude of the vector when determining similarity, so it is less biased towards training data that appears very frequently. The dot product includes the magnitude inherently, so it will tend to value more popular data. Generally, for high-dimensional vector spaces, vectors tend to converge in distance, so Euclidean distance becomes less reliable for large embedding vectors.

  Read more →
• 

  Knowledge Engineering Environment

  Knowledge Engineering Environment (KEE) is a frame-based development tool for expert systems. It was developed and sold by IntelliCorp, and was first released in 1983. It ran on Lisp machines, and was later ported to Lucid Common Lisp with the CLX library, an X Window System (X11) interface for Common Lisp. This version was available on several different UNIX workstations. On KEE, several extensions were offered: Simkit, a frame-based simulation library KEEconnection, database connection between the frame system and relational databases In KEE, frames are called units. Units are used for both individual instances and classes. Frames have slots and slots have facets. Facets can describe, for example, a slot's expected values, its working value, or its inheritance rule. Slots can have multiple values. Behavior can be implemented using a message passing model. KEE provides an extensive graphical user interface (GUI) to create, browse, and manipulate frames. KEE also includes a frame-based rule system. In the KEE knowledge base, rules are frames. Both forward chaining and backward chaining inference are available. KEE supports non-monotonic reasoning through the concepts of worlds. Worlds allow providing alternative slot-values of frames. Through an assumption-based truth or reason maintenance system, inconsistencies can be detected and analyzed. ActiveImages allows graphical displays to be attached to slots of Units. Typical examples are buttons, dials, graphs, and histograms. The graphics are also implemented as Units via KEEPictures, a frame-based graphics library.

  Read more →
• 

  Hyper basis function network

  In machine learning, a Hyper basis function network, or HyperBF network, is a generalization of radial basis function (RBF) networks concept, where the Mahalanobis-like distance is used instead of the Euclidean distance measure. Hyper basis function networks were first introduced by Poggio and Girosi in the 1990 paper “Networks for Approximation and Learning”. == Network Architecture == The typical HyperBF network structure consists of a real input vector x ∈ R n {\displaystyle x\in \mathbb {R} ^{n}} , a hidden layer of activation functions and a linear output layer. The output of the network is a scalar function of the input vector, ϕ : R n → R {\displaystyle \phi :\mathbb {R} ^{n}\to \mathbb {R} } , is given by where N {\displaystyle N} is a number of neurons in the hidden layer, μ j {\displaystyle \mu _{j}} and a j {\displaystyle a_{j}} are the center and weight of neuron j {\displaystyle j} . The activation function ρ j ( | | x − μ j | | ) {\displaystyle \rho _{j}(||x-\mu _{j}||)} at the HyperBF network takes the following form where R j {\displaystyle R_{j}} is a positive definite d × d {\displaystyle d\times d} matrix. Depending on the application, the following types of matrices R j {\displaystyle R_{j}} are usually considered R j = 1 2 σ 2 I d × d {\displaystyle R_{j}={\frac {1}{2\sigma ^{2}}}\mathbb {I} _{d\times d}} , where σ > 0 {\displaystyle \sigma >0} . This case corresponds to the regular RBF network. R j = 1 2 σ j 2 I d × d {\displaystyle R_{j}={\frac {1}{2\sigma _{j}^{2}}}\mathbb {I} _{d\times d}} , where σ j > 0 {\displaystyle \sigma _{j}>0} . In this case, the basis functions are radially symmetric, but are scaled with different width. R j = d i a g ( 1 2 σ j 1 2 , . . . , 1 2 σ j z 2 ) I d × d {\displaystyle R_{j}=diag\left({\frac {1}{2\sigma _{j1}^{2}}},...,{\frac {1}{2\sigma _{jz}^{2}}}\right)\mathbb {I} _{d\times d}} , where σ j i > 0 {\displaystyle \sigma _{ji}>0} . Every neuron has an elliptic shape with a varying size. Positive definite matrix, but not diagonal. == Training == Training HyperBF networks involves estimation of weights a j {\displaystyle a_{j}} , shape and centers of neurons R j {\displaystyle R_{j}} and μ j {\displaystyle \mu _{j}} . Poggio and Girosi (1990) describe the training method with moving centers and adaptable neuron shapes. The outline of the method is provided below. Consider the quadratic loss of the network H [ ϕ ∗ ] = ∑ i = 1 N ( y i − ϕ ∗ ( x i ) ) 2 {\displaystyle H[\phi ^{}]=\sum _{i=1}^{N}(y_{i}-\phi ^{}(x_{i}))^{2}} . The following conditions must be satisfied at the optimum: where R j = W T W {\displaystyle R_{j}=W^{T}W} . Then in the gradient descent method the values of a j , μ j , W {\displaystyle a_{j},\mu _{j},W} that minimize H [ ϕ ∗ ] {\displaystyle H[\phi ^{}]} can be found as a stable fixed point of the following dynamic system: where ω {\displaystyle \omega } determines the rate of convergence. Overall, training HyperBF networks can be computationally challenging. Moreover, the high degree of freedom of HyperBF leads to overfitting and poor generalization. However, HyperBF networks have an important advantage that a small number of neurons is enough for learning complex functions.

  Read more →
• 

  Interactive activation and competition networks

  Interactive activation and competition (IAC) networks are artificial neural networks used to model memory and intuitive generalizations. They are made up of nodes or artificial neurons which are arrayed and activated in ways that emulate the behaviors of human memory. The IAC model is used by the parallel distributed processing (PDP) Group and is associated with James L. McClelland and David E. Rumelhart; it is described in detail in their book Explorations in Parallel Distributed Processing: A Handbook of Models, Programs, and Exercises. This model does not contradict any currently known biological data or theories, and its performance is close enough to human performance as to warrant further investigation.

  Read more →
• 

  Inferential theory of learning

  Inferential Theory of Learning (ITL) is an area of machine learning which describes inferential processes performed by learning agents. ITL has been continuously developed by Ryszard S. Michalski, starting in the 1980s. The first known publication of ITL was in 1983. In the ITL learning process is viewed as a search (inference) through hypotheses space guided by a specific goal. The results of learning need to be stored. Stored information will later be used by the learner for future inferences. Inferences are split into multiple categories including conclusive, deduction, and induction. In order for an inference to be considered complete it was required that all categories must be taken into account. This is how the ITL varies from other machine learning theories like Computational Learning Theory and Statistical Learning Theory; which both use singular forms of inference. == Usage == The most relevant published usage of ITL was in scientific journal published in 2012 and used ITL as a way to describe how agent-based learning works. According to the journal "The Inferential Theory of Learning (ITL) provides an elegant way of describing learning processes by agents".

  Read more →
• 

  PyTorch

  PyTorch is an open-source deep learning library, originally developed by Meta Platforms and currently developed with support from the Linux Foundation. The successor to Torch, PyTorch provides a high-level API that builds upon optimised, low-level implementations of deep learning algorithms and architectures, such as the Transformer, or SGD. Notably, this API simplifies model training and inference to a few lines of code. PyTorch allows for automatic parallelization of training and, internally, implements CUDA bindings that speed training further by leveraging GPU resources. PyTorch utilises the tensor as a fundamental data type, similarly to NumPy. Training is facilitated by a reversed automatic differentiation system, Autograd, that constructs a directed acyclic graph of the operations (and their arguments) executed by a model during its forward pass. With a loss, backpropagation is then undertaken. As of 2025, PyTorch remains one of the most popular deep learning libraries, alongside others such as TensorFlow and Keras. It can be installed using Anaconda package managers. A number of commercial deep learning architectures are built on top of PyTorch, including ChatGPT, Tesla Autopilot, Uber's Pyro, and Hugging Face's Transformers. == History == In 2001, Torch was written and released under a GPL. It was a machine-learning library written in C++ and CUDA, supporting methods including neural networks, support vector machines (SVM), hidden Markov models, etc. Around 2010, it was rewritten by Ronan Collobert, Clement Farabet and Koray Kavuckuoglu. This was known as Torch7 or LuaTorch. This was written so that the backend was in C and the frontend was in Lua. In mid-2016, some developers refactored it to decouple the frontend and the backend, with strong influence from torch-autograd and Chainer. In turn, torch-autograd was influenced by HIPS/autograd. Development on Torch7 ceased in 2018 and was subsumed by the PyTorch project. Meta (formerly known as Facebook) operates both PyTorch and Convolutional Architecture for Fast Feature Embedding (Caffe2), but models defined by the two frameworks were mutually incompatible. The Open Neural Network Exchange (ONNX) project was created by Meta and Microsoft in September 2017 to decouple deep learning frameworks from hardware-specific runtimes, allowing models to be converted between frameworks and optimized for execution providers like NVIDIA’s TensorRT. Caffe2 was merged into PyTorch at the end of March 2018. In September 2022, Meta announced that PyTorch would be governed by the independent PyTorch Foundation, a newly created subsidiary of the Linux Foundation. PyTorch 2.0 was released on 15 March 2023, introducing TorchDynamo, a Python-level compiler that makes code run up to two times faster, along with significant improvements in training and inference performance across major cloud platforms. == PyTorch tensors == PyTorch defines a class called Tensor (torch.Tensor) to store and operate on homogeneous multidimensional rectangular arrays of numbers. PyTorch supports various sub-types of multi-dimensional arrays, or Tensors. PyTorch Tensors are similar to NumPy Arrays, but can also be operated on by a CUDA-capable NVIDIA GPU. PyTorch has also been developing support for other GPU platforms, for example, AMD's ROCm and Apple's Metal Framework. == PyTorch neural networks == PyTorch defines a module called nn (torch.nn) to describe neural networks and to support training. This module offers a comprehensive collection of building blocks for neural networks, including various layers and activation functions, enabling the construction of complex models. Networks are built by inheriting from the torch.nn module and defining the sequence of operations in the forward() function. == PyTorch Serialized File Format == Pytorch can save and load models using its own file format, which is a ZIP64 archive containing the model weights in a Python pickle file, and other information such as the byte order. The file extensions .pt and .pth are commonly used for these files. == Example == The following program shows the low-level functionality of the library with a simple example. The following code block defines a neural network with linear layers using the nn module.

  Read more →
• 

  Logic Theorist

  Logic Theorist is a computer program completed in 1956 by Allen Newell, Herbert A. Simon, and Cliff Shaw. It was the first program deliberately engineered to perform automated reasoning, and has been described as "the first artificial intelligence program". Logic Theorist proved 38 of the first 52 theorems in chapter two of Whitehead and Bertrand Russell's Principia Mathematica, and found new and shorter proofs for some of them. == History == In 1955, when Newell and Simon began to work on the Logic Theorist, the field of artificial intelligence did not yet exist; the term "artificial intelligence" would not be coined until the following summer. Simon was a political scientist who had previously studied the way bureaucracies function as well as developing his theory of bounded rationality (for which he would later win the Nobel Memorial Prize in Economic Sciences in 1978). He believed the study of business organizations requires, like artificial intelligence, an insight into the nature of human problem solving and decision making. Simon has stated that when consulting at RAND Corporation in the early 1950s, he saw a printer typing out a map, using ordinary letters and punctuation as symbols. This led him to think that a machine that could manipulate symbols could simulate decision making and possibly even the process of human thought. The program that printed the map had been written by Newell, a RAND scientist studying logistics and organization theory. For Newell, the decisive moment was in 1954 when Oliver Selfridge came to RAND to describe his work on pattern matching. Watching the presentation, Newell suddenly understood how the interaction of simple, programmable units could accomplish complex behavior, including the intelligent behavior of human beings. "It all happened in one afternoon," he would later say. It was a rare moment of scientific epiphany. "I had such a sense of clarity that this was a new path, and one I was going to go down. I haven't had that sensation very many times. I'm pretty skeptical, and so I don't normally go off on a toot, but I did on that one. Completely absorbed in it—without existing with the two or three levels consciousness so that you're working, and aware that you're working, and aware of the consequences and implications, the normal mode of thought. No. Completely absorbed for ten to twelve hours." Newell and Simon began to talk about the possibility of teaching machines to think. Their first project was a program that could prove mathematical theorems like the ones used in Bertrand Russell and Alfred North Whitehead's Principia Mathematica. They enlisted the help of computer programmer Cliff Shaw, also from RAND, to develop the program. (Newell says "Cliff was the genuine computer scientist of the three".) The first version was hand-simulated: they wrote the program onto 3x5 cards and, as Simon recalled:In January 1956, we assembled my wife and three children together with some graduate students. To each member of the group, we gave one of the cards, so that each one became, in effect, a component of the computer program ... Here was nature imitating art imitating nature. They succeeded in showing that the program could successfully prove theorems as well as a talented mathematician. Eventually Shaw was able to run the program on the computer at RAND's Santa Monica facility. In the summer of 1956, John McCarthy, Marvin Minsky, Claude Shannon and Nathan Rochester organized a conference on the subject of what they called "artificial intelligence" (a term coined by McCarthy for the occasion). Newell and Simon proudly presented the group with the Logic Theorist. It was met with a lukewarm reception. Pamela McCorduck writes "the evidence is that nobody save Newell and Simon themselves sensed the long-range significance of what they were doing." Simon confides that "we were probably fairly arrogant about it all" and adds: They didn't want to hear from us, and we sure didn't want to hear from them: we had something to show them! ... In a way it was ironic because we already had done the first example of what they were after; and second, they didn't pay much attention to it. Logic Theorist soon proved 38 of the first 52 theorems in chapter 2 of the Principia Mathematica. The proof of theorem 2.85 was actually more elegant than the proof produced laboriously by hand by Russell and Whitehead (2026-03-20: What is called here Theorem 2.85 is, in fact, numbered as 2.53 in the page 107 of the 1963 Cambridge University Press edition (https://www.uhu.es/francisco.moreno/gii_mac/docs/Principia_Mathematica_vol1.pdf) and which appears, under the same 2.53 number, on page 112 of the 1910 CUP Edition, according to the digitalization on wikibooks (https://en.wikisource.org/wiki/Russell_%26_Whitehead%27s_Principia_Mathematica/Part_1/Section_A#Discussion_2)). Simon was able to show the new proof to Russell himself who "responded with delight". They attempted to publish the new proof in The Journal of Symbolic Logic, but it was rejected on the grounds that a new proof of an elementary mathematical theorem was not notable, apparently overlooking the fact that one of the authors was a computer program. Newell and Simon formed a lasting partnership, founding one of the first AI laboratories at the Carnegie Institute of Technology and developing a series of influential artificial intelligence programs and ideas, including the General Problem Solver, Soar, and their unified theory of cognition. == Architecture == The Logic Theorist is a program that performs logical processes on logical expressions. The Logic Theorist operates on the following principles: === Expressions === An expression is made of elements. There are two kinds of memories: working and storage. Each working memory contains a single element. The Logic Theorist usually uses 1 to 3 working memories. Each storage memory is a list representing a full expression or a set of elements. In particular, it contains all the axioms and proven logical theorems. An expression is an abstract syntax tree, each node being an element with up to 11 attributes. For example, the logical expression ¬ P → ( Q ∧ ¬ P ) {\displaystyle \neg P\to (Q\wedge \neg P)} is represented as a tree with a root element representing → {\displaystyle \to } . Among the attributes of the root element are pointers to the two elements representing the subexpressions ¬ P {\displaystyle \neg P} and Q ∧ ¬ P {\displaystyle Q\wedge \neg P} . === Processes === There are four kinds of processes, from the lowest to the highest level. Instruction: These are similar to assembly code. They may either perform a primitive operation on an expression in working memory, or perform a conditional jump to another instruction. An example is "put the right sub-element of working-memory 1 to working-memory 2" Elementary process: These are similar to subroutines. A sequence of instructions that can be called. Method: A sequence of elementary processes. There are 4 methods: substitution: given an expression, it attempts to transform it to a proven theorem or axiom by substitutions of variables and logical connectives. detachment: given expression B {\displaystyle B} , it attempts to find a proven theorem or axiom of form A → B ′ {\displaystyle A\to B'} , where B ′ {\displaystyle B'} yields B {\displaystyle B} after substitution, then attempts to prove A {\displaystyle A} by substitution. chaining forward: given expression A → C {\displaystyle A\to C} , it attempts to find for a proven theorem or axiom of form A → B {\displaystyle A\to B} , then attempt to prove B → C {\displaystyle B\to C} by substitution. chaining backward: given expression A → C {\displaystyle A\to C} , it attempts to find for a proven theorem or axiom of form B → C {\displaystyle B\to C} , then attempt to prove A → B {\displaystyle A\to B} by substitution. executive control method: This method applies each of the 4 methods in sequence to each theorem to be proved. == Logic Theorist's influence on AI == Logic Theorist introduced several concepts that would be central to AI research: Reasoning as search Logic Theorist explored a search tree: the root was the initial hypothesis, each branch was a deduction based on the rules of logic. Somewhere in the tree was the goal: the proposition the program intended to prove. The pathway along the branches that led to the goal was a proof – a series of statements, each deduced using the rules of logic, that led from the hypothesis to the proposition to be proved. Heuristics Newell and Simon realized that the search tree would grow exponentially and that they needed to "trim" some branches, using "rules of thumb" to determine which pathways were unlikely to lead to a solution. They called these ad hoc rules "heuristics", using a term introduced by George Pólya in his classic book on mathematical proof, How to Solve It. (Newell had taken courses from Pólya at Stanford). Heuristics would become an important area o

  Read more →
• 

  Philosophy of information

  The philosophy of information (PI) is a branch of philosophy that studies topics relevant to information processing, representational system and consciousness, cognitive science, computer science, information science and information technology. It includes: the critical investigation of the conceptual nature and basic principles of information, including its dynamics, utilisation and sciences the elaboration and application of information-theoretic and computational methodologies to philosophical problems. == History == The philosophy of information (PI) has evolved from the philosophy of artificial intelligence, logic of information, cybernetics, social theory, ethics and the study of language and information. === Logic of information === The logic of information, also known as the logical theory of information, considers the information content of logical signs and expressions along the lines initially developed by Charles Sanders Peirce. === Study of language and information === Later contributions to the field were made by Fred Dretske, Jon Barwise, Brian Cantwell Smith, and others. The Center for the Study of Language and Information (CSLI) was founded at Stanford University in 1983 by philosophers, computer scientists, linguists, and psychologists, under the direction of John Perry and Jon Barwise. === P.I. === More recently this field has become known as the philosophy of information. The expression was coined in the 1990s by Luciano Floridi, who has published prolifically in this area with the intention of elaborating a unified and coherent, conceptual frame for the whole subject. == Definitions of "information" == The concept information has been defined by several theorists. Charles S. Peirce's theory of information was embedded in his wider theory of symbolic communication he called the semiotic, now a major part of semiotics. For Peirce, information integrates the aspects of signs and expressions separately covered by the concepts of denotation and extension, on the one hand, and by connotation and comprehension on the other. Donald M. MacKay says that information is a distinction that makes a difference. According to Luciano Floridi, four kinds of mutually compatible phenomena are commonly referred to as "information": Information about something (e.g. a train timetable) Information as something (e.g. DNA, or fingerprints) Information for something (e.g. algorithms or instructions) Information in something (e.g. a pattern or a constraint). == Philosophical directions == === Computing and philosophy === Recent creative advances and efforts in computing, such as semantic web, ontology engineering, knowledge engineering, and modern artificial intelligence provide philosophy with fertile ideas, new and evolving subject matters, methodologies, and models for philosophical inquiry. While computer science brings new opportunities and challenges to traditional philosophical studies, and changes the ways philosophers understand foundational concepts in philosophy, further major progress in computer science would only be feasible when philosophy provides sound foundations for areas such as bioinformatics, software engineering, knowledge engineering, and ontologies. Classical topics in philosophy, namely, mind, consciousness, experience, reasoning, knowledge, truth, morality and creativity are rapidly becoming common concerns and foci of investigation in computer science, e.g., in areas such as agent computing, software agents, and intelligent mobile agent technologies. According to Luciano Floridi " one can think of several ways for applying computational methods towards philosophical matters: Conceptual experiments in silico: As an innovative extension of an ancient tradition of thought experiment, a trend has begun in philosophy to apply computational modeling schemes to questions in logic, epistemology, philosophy of science, philosophy of biology, philosophy of mind, and so on. Pancomputationalism: On this view, computational and informational concepts are considered to be so powerful that given the right level of abstraction, anything in the world could be modeled and represented as a computational system, and any process could be simulated computationally. Then, however, pancomputationalists have the hard task of providing credible answers to the following two questions: how can one avoid blurring all differences among systems? what would it mean for the system under investigation not to be an informational system (or a computational system, if computation is the same as information processing)?

  Read more →
• 

  Speculative decoding

  Speculative decoding is an inference-time optimization for autoregressive large language models (LLMs) that generates multiple tokens per decoding step instead of one. A smaller draft model proposes a sequence of candidate tokens, and the larger target model verifies them in a single forward pass through a modified rejection sampling scheme. The verification preserves the target model's original output distribution, so the technique produces the same results as standard decoding while cutting latency by roughly two to three times. The name is an analogy to speculative execution in CPU design, where a processor runs instructions along a predicted branch before the outcome is known. == Background == Standard autoregressive decoding in large language models generates one token at a time. The model computes a probability distribution over its vocabulary, samples the next token, and feeds that token back as input. For large models, this process is bottlenecked by memory bandwidth rather than arithmetic throughput: loading the model's parameters from high-bandwidth memory (HBM) to the processor takes up most of the wall-clock time at each step. Because of this, a forward pass over one token and a forward pass over several tokens in a batch take roughly the same time. Speculative decoding relies on this property. == Mechanism == The technique alternates between two phases: drafting and verification. During drafting, a fast approximation model generates a short run of K candidate tokens, typically between 3 and 12. The draft model is usually a much smaller version of the target model or a lightweight auxiliary network. During verification, the target model scores the entire draft sequence in one batched forward pass. A modified rejection sampling algorithm compares the draft and target probabilities at each position. If the target model would have been at least as likely to produce a given token, that token is accepted; the first token that fails is resampled from a corrected distribution, and everything after it is thrown out. The result is that the output distribution is the same as if each token had been generated one at a time. How many tokens get accepted per cycle depends on how well the draft model matches the target. For common words and predictable continuations the match tends to be good, so the target model can confirm several tokens at once. == History == An early precursor was blockwise parallel decoding, proposed in 2018 by Stern, Shazeer, and Uszkoreit. Their method predicted multiple future tokens through auxiliary prediction heads and validated them against the autoregressive model, but it only worked with greedy decoding and did not preserve the full sampling distribution. The modern form of the technique came from Yaniv Leviathan, Matan Kalman, and Yossi Matias at Google Research, who posted "Fast Inference from Transformers via Speculative Decoding" on arXiv in November 2022. Separately and at about the same time, Charlie Chen and colleagues at DeepMind arrived at a closely related method they called speculative sampling, published in February 2023. Both papers introduced the use of rejection sampling to guarantee that the output distribution is unchanged. Leviathan et al. showed roughly 2–3x speedup on T5-XXL (11 billion parameters); Chen et al. reported 2–2.5x on the Chinchilla model (70 billion parameters). The Leviathan et al. paper was presented as an oral at the International Conference on Machine Learning in July 2023. == Variants == SpecInfer (Miao et al., 2024) uses multiple small language models to jointly build a tree of candidate continuations rather than a single chain. The target model verifies the whole tree in parallel and keeps the longest valid path, with reported speedups of 1.5–3.5x. Medusa (Cai et al., 2024) takes a different approach by not using a separate draft model at all. Extra lightweight decoding heads are attached to the target model itself, and each one predicts a token at a different future position. The candidates are evaluated through a tree-structured attention mechanism. The authors measured 2.2–3.6x speedup. EAGLE (Li et al., 2024) performs autoregression on the target model's internal feature representations (specifically the second-to-top layer) rather than on tokens directly. On LLaMA 2 Chat 70B, this gave a 2.7–3.5x latency reduction. Later versions added dynamic draft trees (EAGLE-2) and further optimizations (EAGLE-3), reaching 3–6.5x speedup. == Adoption == By 2024, speculative decoding had become a standard part of production LLM serving. Google uses it in the AI Overviews feature of Google Search. Open-source inference frameworks such as vLLM, NVIDIA's TensorRT-LLM, and SGLang all include built-in support for speculative decoding and its variants. Apple, AWS, and Meta have also published research extending the method or deploying it at scale.

  Read more →
• 

  DeepSeek

  Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd., doing business as DeepSeek, is a Chinese artificial intelligence (AI) company that develops large language models (LLMs). Based in Hangzhou, Zhejiang, DeepSeek is owned and funded by High-Flyer, a Chinese hedge fund. DeepSeek was founded in July 2023 by Liang Wenfeng, the co-founder of High-Flyer, who also serves as the CEO for both of the companies. The company launched an eponymous chatbot alongside its DeepSeek-R1 model in January 2025. DeepSeek-R1 provided responses comparable to other contemporary large language models, such as OpenAI's GPT-4 and o1. Its training cost was reported to be significantly lower than other LLMs. The company claims that it trained its V3 model for US$6 million—far less than the US$100 million cost for OpenAI's GPT-4 in 2023—and using approximately one-tenth the computing power consumed by Meta's comparable model, Llama 3.1. DeepSeek's success against larger and more established rivals has been described as "upending AI". DeepSeek's models are described as "open-weight", meaning the exact parameters are openly shared, but the training data is not openly licensed. Since the January 2025 debut of DeepSeek-R1, the company has made its new models available under free and open-source software licenses, primarily the MIT License. The company reportedly recruits AI researchers from top Chinese universities and also hires from outside traditional computer science fields to broaden its models' knowledge and capabilities. DeepSeek significantly reduced training expenses for their R1 model by incorporating techniques such as mixture of experts (MoE) layers. The company also trained its models during ongoing trade restrictions on AI chip exports to China, using weaker AI chips intended for export and employing fewer units overall. Observers say this breakthrough sent "shock waves" through the industry which were described as triggering a "Sputnik moment" for the US in the field of artificial intelligence, particularly due to its open-source, cost-effective, and high-performing AI models. This threatened established AI hardware leaders such as Nvidia; Nvidia's share price dropped sharply, losing US$600 billion in market value, the largest single-company decline in U.S. stock market history. == History == === Founding and early years (2016–2023) === In February 2016, High-Flyer was co-founded by AI enthusiast Liang Wenfeng, who had been trading since the 2008 financial crisis while attending Zhejiang University. The company began stock trading using a GPU-dependent deep learning model on 21 October 2016; before then, it had used CPU-based linear models. By the end of 2017, most of its trading was driven by AI. Liang established High-Flyer as a hedge fund focused on developing and using AI trading algorithms, and by 2021 the firm was using AI exclusively, often using Nvidia chips. In 2019, the company began constructing its first computing cluster, Fire-Flyer, at a cost of 200 million yuan; it contained 1,100 GPUs interconnected at 200 Gbit/s and was retired after 1.5 years in operation. By 2021, Liang had started buying large quantities of Nvidia GPUs for an AI project, reportedly obtaining 10,000 Nvidia A100 GPUs before the United States restricted chip sales to China. Computing cluster Fire-Flyer 2 began construction in 2021 with a budget of 1 billion yuan. It was reported that in 2022, Fire-Flyer 2's capacity had been used at over 96%, totaling 56.74 million GPU hours. 27% was used to support scientific computing outside the company. During 2022, Fire-Flyer 2 had 5,000 PCIe A100 GPUs in 625 nodes, each containing 8 GPUs. At the time, it exclusively used PCIe instead of the DGX version of A100, since at the time the models it trained could fit within a single 40 GB GPU VRAM and so there was no need for the higher bandwidth of DGX (i.e., it required only data parallelism but not model parallelism). Later, it incorporated NVLinks and NCCL (Nvidia Collective Communications Library) to train larger models that required model parallelism. On 14 April 2023, High-Flyer announced the launch of an artificial general intelligence (AGI) research lab, stating that the new lab would focus on developing AI tools unrelated to the firm's financial business. Two months later, on 17 July 2023, that lab was spun off into an independent company, DeepSeek, with High-Flyer as its principal investor and backer. Venture capital investors were reluctant to provide funding, as they considered it unlikely that the venture would be able to quickly generate an "exit". === Model releases since 2023 === DeepSeek released its first model, DeepSeek Coder, on 2 November 2023, followed by the DeepSeek-LLM series on 29 November 2023. In January 2024, it released two DeepSeek-MoE models (Base and Chat), and in April 3 DeepSeek-Math models (Base, Instruct, and RL). DeepSeek-V2 was released in May 2024, followed a month later by the DeepSeek-Coder V2 series. In September 2024, DeepSeek V2.5 was introduced and revised in December. On 20 November 2024, the preview of DeepSeek-R1-Lite became available via chat. In December, DeepSeek-V3-Base and DeepSeek-V3 (chat) were released. On 20 January 2025, DeepSeek launched the DeepSeek chatbot—based on the DeepSeek-R1 model—free for iOS and Android. By 27 January, DeepSeek surpassed ChatGPT as the most downloaded freeware app on the iOS App Store in the United States, triggering an 18% drop in Nvidia's share price. On 24 March 2025, DeepSeek released DeepSeek-V3-0324 under the MIT License. On 28 May 2025, DeepSeek released DeepSeek-R1-0528 under the MIT License. The model has been noted for more tightly following official Chinese Communist Party ideology and censorship in its answers to questions than prior models. On 21 August 2025, DeepSeek released DeepSeek V3.1 under the MIT License. This model features a hybrid architecture with thinking and non-thinking modes. It also surpasses prior models like V3 and R1, by over 40% on certain benchmarks like SWE-bench and Terminal-bench. It was updated to V3.1-Terminus on 22 September 2025. V3.2-Exp was released on 29 September 2025. It uses DeepSeek Sparse Attention, a more efficient attention mechanism based on previous research published in February. DeepSeek-V3.2 was released on 1 December 2025, alongside a DeepSeek-V3.2-Speciale variant that focused on reasoning. In February 2026, Anthropic accused DeepSeek of using thousands of fraudulent accounts to generate millions of conversations with Claude to train its own large language models. In April 2026, investors began speaking with DeepSeek for a $300 million funding round, which would bring DeepSeek to a total valuation of $10 billion. On April 24, 2026, DeepSeek released a preview of its V4 series, including the 1.6-trillion parameter DeepSeek-V4-Pro and the 284-billion parameter DeepSeek-V4-Flash, both featuring a 1-million token context window, under the MIT License. DeepSeek's V4 LLM has been adopted by key semiconductor manufacturers and artificial intelligence chipmakers such as Huawei and Cambricon. == Company operation == DeepSeek is headquartered in Hangzhou, Zhejiang, and is owned and funded by High-Flyer. Its co-founder, Liang Wenfeng, serves as CEO. As of May 2024, Liang personally held an 84% stake in DeepSeek through two shell corporations. === Strategy === DeepSeek has stated that it focuses on research and does not have immediate plans for commercialization. This posture also means it can skirt certain provisions of China's AI regulations aimed at consumer-facing technologies. DeepSeek's hiring approach emphasizes skills over lengthy work experience, resulting in many hires fresh out of university. The company likewise recruits individuals without computer science backgrounds to expand the range of expertise incorporated into the models, for instance in poetry or advanced mathematics. According to The New York Times, dozens of DeepSeek researchers have or have previously had affiliations with People's Liberation Army laboratories and the Seven Sons of National Defence. Due to the impact of United States restrictions on chips, DeepSeek refined its algorithms to maximise computational efficiency and thereby leveraged older hardware and reduced energy consumption. DeepSeek also expanded on the African continent as it offers more affordable and less power-hungry AI solutions. The company has bolstered African language models and generated a number of startups, for example in Nairobi. Along with Huawei's storage and cloud computing services, the impact on the tech scene in sub-saharan Africa is considerable. DeepSeek offers local data sovereignty and more flexibility compared to Western AI platforms. == Training framework == High-Flyer/DeepSeek had operated at least two primary computing clusters: Fire-Flyer (萤火一号) and Fire-Flyer 2 (萤火二号). Fire-Flyer 1 was constructed in 2019 and was retired after 1.5 years of operation. Fi

  Read more →
• 

  David Krueger (professor)

  David Krueger is an American machine learning professor and advocate for the reduction of risks related to artificial intelligence. Krueger is an assistant professor in Robust, Reasoning, and Responsible AI at the University of Montreal and a Core Academic Member at Mila. == Early life and education == Krueger obtained a B.A. in mathematics from Reed College, and completed his MSc and Ph.D. in Computer Science at the University of Montreal. He trained in deep learning under Yoshua Bengio, Roland Memisevic, and Aaron Courville from 2013 to 2021. Krueger was also an intern on Google DeepMind's AI Safety team in 2018. == Career == Krueger researches deep learning, AI alignment, and AI safety. His work is focused on reducing the risk of human extinction resulting from out-of-control AI systems. Krueger was an assistant professor at the University of Cambridge from 2021 to 2024, before taking a faculty position at the University of Montreal in 2024. In 2023, he was a founding research director at the UK AI Security Institute. That same year, Krueger initiated the Statement on AI Risk, which argues that AI could cause human extinction and was signed by Anthropic's Dario Amodei, OpenAI's Sam Altman, AI expert Geoffrey Hinton, and other leaders. In April 2026, Krueger discussed the risks of advanced AI at a Capitol Hill event hosted by Senator Bernie Sanders. === Evitable === In 2025, Krueger founded Evitable, a nonprofit organization that advocates for an AI moratorium. == Views == Krueger argues that AI will lead to a "gradual disempowerment" of workers, likening AI chips to nuclear bombs. He also says the military use of AI "poses an existential risk to humanity."

  Read more →