EfficientNet

EfficientNet is a family of convolutional neural networks (CNNs) for computer vision published by researchers at Google AI in 2019. Its key innovation is compound scaling, which uniformly scales all dimensions of depth, width, and resolution using a single parameter. EfficientNet models have been adopted in various computer vision tasks, including image classification, object detection, and segmentation. == Compound scaling == EfficientNet introduces compound scaling, which, instead of scaling one dimension of the network at a time, such as depth (number of layers), width (number of channels), or resolution (input image size), uses a compound coefficient ϕ {\displaystyle \phi } to scale all three dimensions simultaneously. Specifically, given a baseline network, the depth, width, and resolution are scaled according to the following equations: depth multiplier: d = α ϕ width multiplier: w = β ϕ resolution multiplier: r = γ ϕ {\displaystyle {\begin{aligned}{\text{depth multiplier: }}d&=\alpha ^{\phi }\\{\text{width multiplier: }}w&=\beta ^{\phi }\\{\text{resolution multiplier: }}r&=\gamma ^{\phi }\end{aligned}}} subject to α ⋅ β 2 ⋅ γ 2 ≈ 2 {\displaystyle \alpha \cdot \beta ^{2}\cdot \gamma ^{2}\approx 2} and α ≥ 1 , β ≥ 1 , γ ≥ 1 {\displaystyle \alpha \geq 1,\beta \geq 1,\gamma \geq 1} . The α ⋅ β 2 ⋅ γ 2 ≈ 2 {\displaystyle \alpha \cdot \beta ^{2}\cdot \gamma ^{2}\approx 2} condition is such that increasing ϕ {\displaystyle \phi } by a factor of ϕ 0 {\displaystyle \phi _{0}} would increase the total FLOPs of running the network on an image approximately 2 ϕ 0 {\displaystyle 2^{\phi _{0}}} times. The hyperparameters α {\displaystyle \alpha } , β {\displaystyle \beta } , and γ {\displaystyle \gamma } are determined by a small grid search. The original paper suggested 1.2, 1.1, and 1.15, respectively. Architecturally, they optimized the choice of modules by neural architecture search (NAS), and found that the inverted bottleneck convolution (which they called MBConv) used in MobileNet worked well. The EfficientNet family is a stack of MBConv layers, with shapes determined by the compound scaling. The original publication consisted of 8 models, from EfficientNet-B0 to EfficientNet-B7, with increasing model size and accuracy. EfficientNet-B0 is the baseline network, and subsequent models are obtained by scaling the baseline network by increasing ϕ {\displaystyle \phi } . == Variants == EfficientNet has been adapted for fast inference on edge TPUs and centralized TPU or GPU clusters by NAS. EfficientNet V2 was published in June 2021. The architecture was improved by further NAS search with more types of convolutional layers. It also introduced a training method, which progressively increases image size during training, and uses regularization techniques like dropout, RandAugment, and Mixup. The authors claim this approach mitigates accuracy drops often associated with progressive resizing.

Computer appliance

A computer appliance is a computer system with a combination of hardware, software, or firmware that is specifically designed to provide a particular computing resource. Such devices became known as appliances because of the similarity in role or management to a home appliance, which are generally closed and sealed, and are not serviceable by the user or owner. The hardware and software are delivered as an integrated product and may even be pre-configured before delivery to a customer, to provide a turn-key solution for a particular application. Unlike general purpose computers, appliances are generally not designed to allow the customers to change the software and the underlying operating system, or to flexibly reconfigure the hardware. Another form of appliance is the virtual appliance, which has similar functionality to a dedicated hardware appliance, but is distributed as a software virtual machine image for a hypervisor-equipped device. == Overview == Traditionally, software applications run on top of a general-purpose operating system, which uses the hardware resources of the computer (primarily memory, disk storage, processing power, and networking bandwidth) to meet the computing needs of the user. The main issue with the traditional model is related to complexity. It is complex to integrate the operating system and applications with a hardware platform, and complex to support it afterwards. By tightly constraining the variations of the hardware and software, the appliance becomes easily deployable, and can be used without nearly as wide (or deep) IT knowledge. Additionally, when problems and errors appear, the supporting staff very rarely needs to explore them deeply to understand the matter thoroughly. The staff needs merely training on the appliance management software to be able to resolve most of problems. In all forms of the computer appliance model, customers benefit from easy operations. The appliance has exactly one combination of hardware and operating system and application software, which has been pre-installed at the factory. This prevents customers from needing to perform complex integration work, and dramatically simplifies troubleshooting. In fact, this "turnkey operation" characteristic is the driving benefit that customers seek when purchasing appliances. To be considered an appliance, the (hardware) device needs to be integrated with software, and both are supplied as a package. This distinguishes appliances from "home grown" solutions, or solutions requiring complex implementations by integrators or value-added resellers (VARs). The appliance approach helps to decouple the various systems and applications, for example in the data center. Once a resource is decoupled, in theory it can be also centralized to become shared among many systems, centrally managed and optimized, all without requiring changes to any other system. == Tradeoffs of the computer appliance approach == The major disadvantage of deploying a computer appliance is that since they are designed to supply a specific resource, they most often include a customized operating system running over specialized hardware, neither of which are likely to be compatible with the other systems previously deployed. Customers lose flexibility. One may believe that a proprietary embedded operating system, or operating system within an application, can make the appliance much more secure from common cyber attacks. However, the opposite is true. Security by obscurity is a poor security decision, and appliances are often plagued by security issues as evidenced by the proliferation of IoT devices. == Types of appliances == The variety of computer appliances reflects the wide range of computing resources they provide to applications. Some examples: Storage appliances provide large amounts of storage, often available to many machines on the network. See Network-attached storage and Storage area network. Network appliances are general purpose routers which may also provide firewall protection, Transport Layer Security (TLS), messaging, access to specialized networking protocols (like the ebXML Message Service) and bandwidth multiplexing for the multiple systems they front-end. Backup and disaster recovery appliances computer appliances that are integrated backup software and backup targets, sometimes with hypervisors to support local DR of protected servers. They are often a gateway to a full DRaaS solution. Firewall and Security appliances Dedicated network appliances that are designed to protect computer networks from unwanted traffic. IIoT and MES Gateway appliances Computer appliances that are designed to translate data bidirectionally between control systems and enterprise systems. Proprietary, embedded, firmware applications running on the appliance use point-to-point connections to translate data between field devices in their native automation protocols and MES systems through their APIs, ODBC, or RESTful interfaces. Anti-spam appliances for e-mail spam Software appliances A single application server appliance, with just enough operating system (JeOS) for it to run. Virtual machine appliances consist of a "hypervisor style" embedded operating system running on appliance hardware. The hypervisor layer is matched to the hardware of the appliance, and cannot be varied by the customer, but the customer may load other operating systems and applications onto the appliance in the form of virtual machines. == Consumer appliances == Aside from its deployment within data centers, many computer appliances are directly used by the general public. These include: Digital video recorder Residential gateway Network-attached storage (NAS) Video game console Consumer uses stress the need for an appliance to have easy installation, configuration, and operation, with little or no technical knowledge being necessary. == Appliances in industrial automation == The world of industrial automation has been rich in appliances. These appliances have been hardened to withstand temperature and vibration extremes. These appliances are also highly configurable, enabling customization to meet a wide variety of applications. The key benefits of an appliance in automation are: Reduced downtime - a failed appliance is typically replaced with a COTS replacement and its task is quickly and easily reloaded from a backup. Highly scalable - appliances are typically targeted solutions for an area of a plant or process. As the requirements change, scalability is achieved through the installation of another appliance. Automation concepts are easily replicated throughout the enterprise by standardizing on appliances to perform the needed tasks, as opposed to the development of custom automation schemes for each task. Low TCO (total cost of ownership) - appliances are developed, tested and supported by automation product vendors and undergo a much broader level of quality testing than custom designed automation solutions. The use of appliances in automation reduce the level of testing needed in each individual application. Reduced design time - appliances perform specific functions and although they are highly configurable, they are typically self documenting. This enables appliance based solutions to be transferred from engineer to engineer with minimal need for training and documentation. Types of automation appliances: PLC (programmable logic controller) - Programmable logic controllers are appliances that are typically used for discrete control and offer a wide range of Input and Output options. They are configured through standardized programming languages such as IEC-1131. PID (proportional–integral–derivative controller) - PID controllers are appliances that monitor a process variable and, based on an error term, effect change on a control output (manipulated variable) to drive the process variable to a setpoint. PAC (programmable automation controller) - Programmable automation controllers are appliances that embody properties of both PLCs and PID controllers enabling the integration of both analog and discrete control. Universal gateway - A universal gateway appliance has the ability to communicate with a variety of devices through their respective communication protocols, and will affect data transactions between them. This in increasingly important as manufacturing strives to improve agility, quality, production rates, production costs and reduce downtime through enhanced M2M (machine to machine) communications. EATMs (Enterprise Appliance Transaction Modules) - Enterprise appliance transaction modules are appliances that affect data transactions from plant floor automation systems to enterprise business systems. They communicate to plant floor equipment through various vendor automation protocols, and communicate to business systems through database communication protocols such as JMS (Java Message Service) and SQL (Structured Query Language). == Internal structure == There are several

Collective operation

Collective operations are building blocks for interaction patterns, that are often used in SPMD algorithms in the parallel programming context. Hence, there is an interest in efficient realizations of these operations. A realization of the collective operations is provided by the Message Passing Interface (MPI). == Definitions == In all asymptotic runtime functions, we denote the latency α {\displaystyle \alpha } (or startup time per message, independent of message size), the communication cost per word β {\displaystyle \beta } , the number of processing units p {\displaystyle p} and the input size per node n {\displaystyle n} . In cases where we have initial messages on more than one node we assume that all local messages are of the same size. To address individual processing units we use p i ∈ { p 0 , p 1 , … , p p − 1 } {\displaystyle p_{i}\in \{p_{0},p_{1},\dots ,p_{p-1}\}} . If we do not have an equal distribution, i.e. node p i {\displaystyle p_{i}} has a message of size n i {\displaystyle n_{i}} , we get an upper bound for the runtime by setting n = max ( n 0 , n 1 , … , n p − 1 ) {\displaystyle n=\max(n_{0},n_{1},\dots ,n_{p-1})} . A distributed memory model is assumed. The concepts are similar for the shared memory model. However, shared memory systems can provide hardware support for some operations like broadcast (§ Broadcast) for example, which allows convenient concurrent read. Thus, new algorithmic possibilities can become available. == Broadcast == The broadcast pattern is used to distribute data from one processing unit to all processing units, which is often needed in SPMD parallel programs to dispense input or global values. Broadcast can be interpreted as an inverse version of the reduce pattern (§ Reduce). Initially only root r {\displaystyle r} with i d {\displaystyle id} 0 {\displaystyle 0} stores message m {\displaystyle m} . During broadcast m {\displaystyle m} is sent to the remaining processing units, so that eventually m {\displaystyle m} is available to all processing units. Since an implementation by means of a sequential for-loop with p − 1 {\displaystyle p-1} iterations becomes a bottleneck, divide-and-conquer approaches are common. One possibility is to utilize a binomial tree structure with the requirement that p {\displaystyle p} has to be a power of two. When a processing unit is responsible for sending m {\displaystyle m} to processing units i . . j {\displaystyle i..j} , it sends m {\displaystyle m} to processing unit ⌈ ( i + j ) / 2 ⌉ {\displaystyle \left\lceil (i+j)/2\right\rceil } and delegates responsibility for the processing units ⌈ ( i + j ) / 2 ⌉ . . j {\displaystyle \left\lceil (i+j)/2\right\rceil ..j} to it, while its own responsibility is cut down to i . . ⌈ ( i + j ) / 2 ⌉ − 1 {\displaystyle i..\left\lceil (i+j)/2\right\rceil -1} . Binomial trees have a problem with long messages m {\displaystyle m} . The receiving unit of m {\displaystyle m} can only propagate the message to other units, after it received the whole message. In the meantime, the communication network is not utilized. Therefore pipelining on binary trees is used, where m {\displaystyle m} is split into an array of k {\displaystyle k} packets of size ⌈ n / k ⌉ {\displaystyle \left\lceil n/k\right\rceil } . The packets are then broadcast one after another, so that data is distributed fast in the communication network. Pipelined broadcast on balanced binary tree is possible in O ( α log ⁡ p + β n ) {\displaystyle {\mathcal {O}}(\alpha \log p+\beta n)} , whereas for the non-pipelined case it takes O ( ( α + β n ) log ⁡ p ) {\displaystyle {\mathcal {O}}((\alpha +\beta n)\log p)} cost. == Reduce == The reduce pattern is used to collect data or partial results from different processing units and to combine them into a global result by a chosen operator. Given p {\displaystyle p} processing units, message m i {\displaystyle m_{i}} is on processing unit p i {\displaystyle p_{i}} initially. All m i {\displaystyle m_{i}} are aggregated by ⊗ {\displaystyle \otimes } and the result is eventually stored on p 0 {\displaystyle p_{0}} . The reduction operator ⊗ {\displaystyle \otimes } must be associative at least. Some algorithms require a commutative operator with a neutral element. Operators like s u m {\displaystyle sum} , m i n {\displaystyle min} , m a x {\displaystyle max} are common. Implementation considerations are similar to broadcast (§ Broadcast). For pipelining on binary trees the message must be representable as a vector of smaller object for component-wise reduction. Pipelined reduce on a balanced binary tree is possible in O ( α log ⁡ p + β n ) {\displaystyle {\mathcal {O}}(\alpha \log p+\beta n)} . == All-Reduce == The all-reduce pattern (also called allreduce) is used if the result of a reduce operation (§ Reduce) must be distributed to all processing units. Given p {\displaystyle p} processing units, message m i {\displaystyle m_{i}} is on processing unit p i {\displaystyle p_{i}} initially. All m i {\displaystyle m_{i}} are aggregated by an operator ⊗ {\displaystyle \otimes } and the result is eventually stored on all p i {\displaystyle p_{i}} . Analog to the reduce operation, the operator ⊗ {\displaystyle \otimes } must be at least associative. All-reduce can be interpreted as a reduce operation with a subsequent broadcast (§ Broadcast). For long messages a corresponding implementation is suitable, whereas for short messages, the latency can be reduced by using a hypercube (Hypercube (communication pattern) § All-Gather/ All-Reduce) topology, if p {\displaystyle p} is a power of two. All-reduce can also be implemented with a butterfly algorithm and achieve optimal latency and bandwidth. All-reduce is possible in O ( α log ⁡ p + β n ) {\displaystyle {\mathcal {O}}(\alpha \log p+\beta n)} , since reduce and broadcast are possible in O ( α log ⁡ p + β n ) {\displaystyle {\mathcal {O}}(\alpha \log p+\beta n)} with pipelining on balanced binary trees. All-reduce implemented with a butterfly algorithm achieves the same asymptotic runtime. == Prefix-Sum/Scan == The prefix-sum or scan operation is used to collect data or partial results from different processing units and to compute intermediate results by an operator, which are stored on those processing units. It can be seen as a generalization of the reduce operation (§ Reduce). Given p {\displaystyle p} processing units, message m i {\displaystyle m_{i}} is on processing unit p i {\displaystyle p_{i}} . The operator ⊗ {\displaystyle \otimes } must be at least associative, whereas some algorithms require also a commutative operator and a neutral element. Common operators are s u m {\displaystyle sum} , m i n {\displaystyle min} and m a x {\displaystyle max} . Eventually processing unit p i {\displaystyle p_{i}} stores the prefix sum ⊗ i ′ <= i {\displaystyle \otimes _{i'<=i}} m i ′ {\displaystyle m_{i'}} . In the case of the so-called exclusive prefix sum, processing unit p i {\displaystyle p_{i}} stores the prefix sum ⊗ i ′ < i {\displaystyle \otimes _{i'

TurboQuant

TurboQuant is an online vector quantization algorithm for compressing high-dimensional Euclidean vectors while preserving their geometric structure. It was proposed in 2025 by Amir Zandieh, Majid Daliri, Majid Hadian, and Vahab Mirrokni in the paper TurboQuant: Online Vector Quantization with Near-optimal Distortion Rate. The paper lists Zandieh and Mirrokni as affiliated with Google Research, Daliri with New York University, and Hadian with Google DeepMind. The method was developed for applications including large language model (LLM) inference, key–value (KV) cache compression, vector databases, and nearest neighbor search. TurboQuant consists of two related algorithms: TurboQuantmse, which is optimized for mean squared error (MSE), and TurboQuantprod, which is optimized for unbiased inner product estimation. The algorithm uses a random rotation of input vectors, applies scalar quantizers to the rotated coordinates, and, for inner-product estimation, applies a one-bit Quantized Johnson–Lindenstrauss (QJL) transform to the residual error. == Background == Vector quantization is a compression method that maps high-dimensional vectors to a finite set of codewords. The problem has roots in Shannon's source coding theory and rate–distortion theory. In machine learning and information retrieval, vector quantization is used to reduce the memory required to store embeddings, activation vectors, and other numerical representations. In Transformer-based large language models, the KV cache stores key and value vectors from previous tokens during autoregressive decoding. The size of this cache grows with context length, the number of attention heads, and the number of concurrent requests, making it a major memory bottleneck in LLM serving. Similar compression problems appear in vector search, where large collections of embedding vectors must be stored and searched efficiently. Earlier approaches to vector quantization include product quantization, scalar quantization, and data-dependent k-means codebook construction. The TurboQuant paper argues that many existing methods either require offline preprocessing and calibration or suffer from suboptimal distortion guarantees in online settings. == Algorithm == === TurboQuantmse === TurboQuantmse is the version of the algorithm optimized for mean-squared error. For a unit vector x ∈ S d − 1 {\displaystyle x\in S^{d-1}} , the algorithm first applies a random rotation matrix Π ∈ R d × d {\displaystyle \Pi \in \mathbb {R} ^{d\times d}} and sets z = Π x {\displaystyle z=\Pi x} . Each coordinate of the rotated vector follows a shifted and scaled beta distribution, which converges to a normal distribution in high dimensions. In high dimensions, distinct coordinates also become nearly independent, allowing the algorithm to apply scalar quantizers independently to each coordinate. The scalar quantizer is constructed by solving a one-dimensional continuous k-means or Lloyd–Max quantization problem. If the centroids are c 1 , c 2 , … , c 2 b {\displaystyle c_{1},c_{2},\ldots ,c_{2^{b}}} , the quantization step stores, for each coordinate, i d x j = ⁡ a r g m i n k ∈ [ 2 b ] | z j − c k | . {\displaystyle \mathrm {idx} _{j}=\operatorname {} {arg\,min}_{k\in [2^{b}]}|z_{j}-c_{k}|.} During dequantization, the stored index for each coordinate is replaced by the corresponding centroid, giving a reconstructed rotated vector z ~ {\displaystyle {\tilde {z}}} . The algorithm then rotates back: x ~ = Π ⊤ z ~ . {\displaystyle {\tilde {x}}=\Pi ^{\top }{\tilde {z}}.} The paper gives the following bound for TurboQuantmse: D m s e ≤ 3 π 2 ⋅ 1 4 b . {\displaystyle D_{\mathrm {mse} }\leq {\frac {\sqrt {3\pi }}{2}}\cdot {\frac {1}{4^{b}}}.} It also reports finer-grained MSE values of approximately 0.36, 0.117, 0.03, and 0.009 for bit-widths b = 1 , 2 , 3 , 4 {\displaystyle b=1,2,3,4} , respectively. === TurboQuantprod === TurboQuantprod is optimized for unbiased inner-product estimation. The authors note that an MSE-optimized quantizer may introduce bias when used to estimate inner products. To address this, TurboQuantprod first applies TurboQuantmse with bit-width b − 1 {\displaystyle b-1} , then applies a one-bit Quantized Johnson–Lindenstrauss transform to the remaining residual vector. Let r = x − Q m s e − 1 ( Q m s e ( x ) ) {\displaystyle r=x-Q_{\mathrm {mse} }^{-1}(Q_{\mathrm {mse} }(x))} be the residual after MSE quantization, and let γ = ‖ r ‖ 2 {\displaystyle \gamma =\|r\|_{2}} . The QJL step stores a sign vector for the residual. For γ ≠ 0 {\displaystyle \gamma \neq 0} , this can be written using the normalized residual u = r / γ {\displaystyle u=r/\gamma } : q j l = sign ⁡ ( S u ) , {\displaystyle qjl=\operatorname {sign} (Su),} where S ∈ R d × d {\displaystyle S\in \mathbb {R} ^{d\times d}} is a random projection matrix. Since the sign function is invariant under positive rescaling, this is equivalent to sign ⁡ ( S r ) {\displaystyle \operatorname {sign} (Sr)} when r ≠ 0 {\displaystyle r\neq 0} . If γ = 0 {\displaystyle \gamma =0} , the residual correction is zero. TurboQuantprod stores the MSE quantization, the QJL sign vector, and the residual norm: Q p r o d ( x ) = [ Q m s e ( x ) , q j l , γ ] . {\displaystyle Q_{\mathrm {prod} }(x)=\left[Q_{\mathrm {mse} }(x),qjl,\gamma \right].} The dequantized vector is reconstructed as x ~ = x ~ m s e + π / 2 d γ S ⊤ q j l . {\displaystyle {\tilde {x}}={\tilde {x}}_{\mathrm {mse} }+{\frac {\sqrt {\pi /2}}{d}}\,\gamma S^{\top }qjl.} The paper proves that TurboQuantprod is unbiased for inner-product estimation: E x ~ [ ⟨ y , x ~ ⟩ ] = ⟨ y , x ⟩ . {\displaystyle \mathbb {E} _{\tilde {x}}\left[\langle y,{\tilde {x}}\rangle \right]=\langle y,x\rangle .} It also gives the distortion bound D p r o d ≤ 3 π 2 ⋅ ‖ y ‖ 2 2 d ⋅ 1 4 b . {\displaystyle D_{\mathrm {prod} }\leq {\frac {\sqrt {3\pi }}{2}}\cdot {\frac {\|y\|_{2}^{2}}{d}}\cdot {\frac {1}{4^{b}}}.} == Performance and applications == The TurboQuant paper reports that the algorithm achieves near-optimal distortion rates within a small constant factor of information-theoretic lower bounds. The authors report that, for KV cache quantization, TurboQuant achieved quality neutrality at 3.5 bits per channel and marginal degradation at 2.5 bits per channel. In long-context LLM experiments using Llama 3.1 8B Instruct, the paper evaluated the method on a "needle-in-a-haystack" retrieval task with document lengths from 4,000 to 104,000 tokens. It reported that TurboQuant matched the uncompressed full-precision baseline while using more than 4× compression, and compared the method against PolarQuant, SnapKV, PyramidKV, and KIVI. Google Research stated that TurboQuant was evaluated on long-context benchmarks including LongBench, Needle in a Haystack, ZeroSCROLLS, RULER, and L-Eval using open-source models including Gemma and Mistral. According to a report in Tom's Hardware, Google described the method as reducing KV-cache memory by at least six times and achieving up to an eightfold improvement in attention-logit computation on Nvidia H100 GPUs compared with unquantized 32-bit keys. TurboQuant has also been applied to nearest-neighbor vector search. The original paper reports experiments on DBpedia entity embeddings and GloVe embeddings, comparing TurboQuant with product quantization and other vector-search quantization baselines. == Relationship to other methods == TurboQuant is related to several methods for efficient large language model inference and high-dimensional search: Product quantization – a vector quantization technique widely used for approximate nearest-neighbor search Quantization (machine learning) – reducing the numerical precision of weights, activations, or cached tensors in machine learning models PagedAttention – a memory-management algorithm for LLM serving that reduces fragmentation in the KV cache Johnson–Lindenstrauss lemma – a result in high-dimensional geometry used in random projection methods Lloyd's algorithm – an algorithm for scalar and vector quantization, including k-means-style codebook construction Unlike PagedAttention, which focuses on memory allocation and cache layout, TurboQuant reduces the numerical storage cost of the vectors themselves. Unlike many product-quantization methods, TurboQuant is designed to be data-oblivious and online, avoiding dataset-specific codebook training. == Limitations == The strongest performance claims for TurboQuant come from the original paper and Google Research's own publication. Coverage in technology media has noted that the broader impact of the method will depend on real-world implementation details, workloads, and hardware architectures.

Sikidy

Sikidy is a form of algebraic geomancy practiced by Malagasy peoples in Madagascar. It involves algorithmic operations performed on random data generated from tree seeds, which are ritually arranged in a tableau called a toetry and divinely interpreted after being mathematically operated on. Columns of seeds, designated "slaves" or "princes" belonging to respective "lands" for each, interact symbolically to express vintana ('fate') in the interpretation of the diviner. The diviner also prescribes solutions to problems and ways to avoid fated misfortune, often involving a sacrifice. The centuries-old practice derives from Islamic influence brought to the island by medieval Arab traders. The sikidy is consulted for a range of divinatory questions pertaining to fate and the future, including identifying sources of and rectifying misfortune, reading the fate of newborns, and planning annual migrations. The mathematics of sikidy involves Boolean algebra, symbolic logic and parity. == History == The practice is several centuries old, and is influenced by Arab geomantic traditions of Arab Muslim traders on the island. Most writers link the origins of sikidy to the "sea-going trade involving the southwest coast of India, the Persian Gulf, and the east coast of Africa in the 9th or 10th century C.E." Stephen Ellis and Solofo Randrianja describe sikidy as "probably one of the oldest components of Malagasy culture", writing that it most likely the product of an indigenous divinatory art later influenced by Islamic practice. Umar H. D. Danfulani writes that the integration of Arabic divination into indigenous divination is "clearly demonstrated" in Madagascar, where the Arabic astrological system was adapted to the indigenous agricultural system and meshed with Malagasy lunar months by "adapting indigenous months, volana, to the astrological months, vintana". Danfulani also describes the concepts in sikidy of "houses" (lands) and "kings in their houses" as retained from medieval Arabic astrology. Chemillier et al. say the practice's spread across Madagascar likely originated with the southeastern Antemoro people, among whom Arab influence was the strongest. Though the etymology of sikidy is unknown, it has been posited that the word derives from the Arabic sichr ('incantation' or 'charm'). Sikidy was of central importance to pre-Christian Malagasy religion, with one practitioner quoted in 1892 as calling sikidy "the Bible of our ancestors". A missionary report from 1616 describes one form of sikidy using tamarind seeds, and another using fingered markings in the sand. The early colonial French governor of Madagascar Étienne de Flacourt documented sikidy in the mid-17th century: Matatane country in southeastern Madagascar [...] where the Antemoro [...] live was a center of astrological study as early as the fourteenth century [...]. This area was also the site of early Arab settlements, although strict Islamic observances were lost centuries ago [...]. Historical evidence shows that Antemoro diviners, bearers of the astrological system, infiltrated nearly all the ancient kingdoms of Madagascar beginning in the sixteenth century. [...] Today, although many persons claim to be ombiasy [diviners], only the Antemoro diviners are considered true professionals. The area is still a famous place of learning where specialists go for training and then return to their home communities with a certain body of knowledge. Now we can better understand the degree of similarity of divination forms found throughout Madagascar. For centuries Matitanana has remained a training center for diviners who have migrated widely, usually attaining important positions in their home communities and with various royal families. Comparison of contemporary rites with centuries-old texts show that sikidy has been remarkably unchanged throughout its history. The "infiltration" of Malagasy kingdoms by Antemoro diviners, and Matitanana's role as a place for astrological and divinatory learning, help to explain the relatively uniform practicing of sikidy across Madagascar. Chemallier et al. write that the mathematical construction of the arrangement of seeds is procedurally consistent across all of Madagascar, with variations in practice between groups and regions being limited to more minor aspects, such as the alignment of figures according to cardinal directions. One exception is the simplified Merina sikidy joria. === Origin myths === Mythic tradition relating to the origin of sikidy "links [the practice] both to the return by walking on water of Arab ancestors who had intermarried with Malagasy but then left, and to the names of the days of the week" and holds that the art was supernaturally communicated to the ancestors, with Zanahary (the supreme deity of Malagasy religion) giving it to Ranakandriana, who then gave it to a line of diviners (Ranakandriana to Ramanitralanana to Rabibi-andrano to Andriambavi-maitso (who was a woman) to Andriam-bavi-nosy), the last of whom terminated the monopoly by giving it to the people, declaring: "Behold, I give you the sikidy, of which you may inquire what offerings you should present in order to obtain blessings; and what expiation you should make so as to avert evils, when any are ill or under apprehension of some future calamity". A mythic anecdote of Ranakandriana says that two men observed him one day playing in the sand. In fact he was practicing a form of sikidy worked in sand called sikidy alanana. The two men seized him, and Ranakandriana promised that he would teach them something if they released him. They agreed, and Ranakandriana taught them in depth how to work the sikidy. The two men then went to their chief and told him that they could tell him "the past and the future—what was good and what was bad—what increased and what diminished." The chief asked them to tell him how he could obtain plenty of cattle. The two men worked their sikidy and told the chief to kill all of his bulls, and that "great numbers would come to him" on the following Friday. The chieftain, doubting, asked what would happen if their prediction didn't come true, and the two men promised they would pay with their lives. The chief agreed and killed his bulls. On Thursday, thinking he'd been duped, he prematurely killed the first man of the two who'd told him about the divinatory art. On Friday, however, "vast herds" came amidst heavy rain, actually filling an immense plain in their crowd. The chieftain lamented the mpisikidy's wrongful execution and ordered for him a pompous funeral. The chieftain took the second man as his close adviser and friend, and trusted the sikidy forever afterwards. The British missionary William Ellis recorded in 1839 two idiomatic expressions used in Madagascar that come from this story: "Tsy mahandry andro Zoma" (lit. 'He cannot wait 'til Friday') is said of someone extremely impatient, and heavy rainshowers falling in rapid succession are called "sese omby" (lit. 'a crowding together of cattle'). == Rites and arrangement of seeds == The divination is performed by a practitioner called an mpisikidy, ny màsina (lit. 'sacred one'), ombiasy, or ambiàsa (derived from the Arabic anbia, meaning 'prophet') who guides the client through the process and interprets the results in the context of the client's inquiries and desires. As part of an mpisikidy's formal initiation into the art, which includes a long period of apprenticeship, the initiate (called a mianatsy) must gather 124 and 200 fàno (Entada sp.) or kily (tamarind) tree seeds for his subsequent ritual use in sikidy. Raymond Decary writes that, at least among the Sakalava, a man must be 40 years old before learning and practicing sikidy, or he risks death. Before beginning to study, a student practitioner must make incisions at the tips of his index finger, his middle finger, and his tongue, and put within the incisions a paste containing red pepper and crushed wasp. This paste impregnates the fingers that will move the seeds of the sikidy and the tongue that will speak their revelations with the power to decipher the sikidy. Once this is done, he leaves at dawn to search for a fano (Entada chrysostachys) tree. Upon finding it, he throws his spear at its branches, shaking the tree and causing its large seed pods to fall. During this act, some initiates say: "When you were on the steep peak and in the dense forest, on you the crabs climbed, from you the crocodiles made their bed, with their paws the birds trod on you. Whether you are suspended in the trees or buried, you are never dried up nor rotten." In his study (written in 1941 and revised in 1948), Decary reported that the salary paid by a mianatsy to his master is "not very high": up to five francs, plus a red rooster's feather. The mpisikidy ritually arranges his seeds into a sixteen-column table consisting of four columns of randomly-generated data (representing fate) and eight columns of data derived from logical ope

Qloo

Qloo (pronounced "clue") is a company that uses artificial intelligence (AI) to understand taste and cultural correlations. It provides companies with an application programming interface (API). It received funding from Leonardo DiCaprio, Elton John, Barry Sternlicht, Pierre Lagrange and others. Qloo establishes consumer preference correlations via machine learning across data spanning cultural domains including music, film, television, dining, nightlife, fashion, books, and travel. The recommender system uses AI to predict correlations for further applications. == History == Qloo was founded in 2012 by chief executive officer Alex Elias and chief operating officer Jay Alger. Qloo initially launched an app designed for consumers, allowing them to understand their own tastes and receive personalized recommendations. The company amassed several million users and built a large catalog of cultural entities and corresponding user sentiment. In 2012, Qloo raised $1.4 million in seed funding from investors including Cedric the Entertainer, and venture capital firm Kindler Capital. Qloo had a public beta release in November 2012 after its initial funding. In 2013, the company raised an additional $1.6 million from Cross Creek Pictures founding partner Tommy Thompson, and Samih Toukan and Hussam Khoury, founders of Maktoob, an Internet services company purchased by Yahoo! for $164 million in 2009. On November 14, 2013, a website and an iPhone app were announced. The company later released an Android app, and tablet versions, in mid-2014. In 2015, Twitter approached Qloo about powering personalized social feeds and targeted eCommerce ads on the platform based on what users were posting. Qloo developed an enterprise-grade API to support Twitter’s needs. Twitter ended up pivoting to enable brands to use the social platform for customer service and support, but Qloo was able to sell access to its cultural intelligence via API to many other enterprise clients, marking the official transition from a B2C company to a B2B company. In 2016, Qloo secured $4.5 million in venture capital investment. The $4.5 million was split between a number of investors, including Barry Sternlicht, Pierre Lagrange, and Leonardo DiCaprio. In July 2017, Qloo raised $6.5 million in funding rounds from AXA Strategic Ventures, and Elton John. Following the investment, the founders stated in an interview with Tech Crunch that they would use the investment to expand Qloo's database. They hoped the move would secure larger contracts with corporate clients. At the time, clients already included Fortune 500 companies such as Twitter, PepsiCo, and BMW. In 2019, the company announced that it had acquired cultural recommendation service TasteDive, with Alex Elias becoming chairman of TasteDive. In September 2019, Qloo was named among the Top 14 Artificial Intelligence APIs by ProgrammableWeb. In 2022, Qloo raised $15M in Series B funding from Eldridge and AXA Venture Partners, enabling the privacy-centric AI leader to expand its team of world-class data scientists, enrich its technology, and build on its sales channels in order to continue to offer premier insights into global consumer taste for Fortune 500 companies across the globe. Qloo was recognized as the "Best Decision Intelligence Company" at the 2023 AI Breakthrough Awards. Also in 2023, the company was awarded a Top Performer Award by SourceForge. As of 2024, Qloo is a three-time Inc. 5000 honoree: No. 360 (2022), No. 344 (2021), No. 187 (2020). Qloo raised $25 million Series C round on February 21, 2024. The round was led by AI Ventures with participation from AXA Venture Partners, Eldridge, and Moderne Ventures, allowing Qloo to address new commercial surface areas for Taste AI, including on-device learning and foundational models leveraging Qloo, as well as introduce self-service platform to make consumer and taste analytics available to small and mid-sized enterprises and individuals. Qloo also announced pursuing opportunistic M&A using its balance sheet along the lines of the TasteDive acquisition completed, which expanded Qloo's first-party data moat and corpus of cultural learning. This latest financing brought the total amount raised since the company's founding in 2012 to over $56 million. == Services and features == Qloo calls itself a cultural AI platform to provide real-time correlation data across domains of culture and entertainment including: film, music, television, dining, nightlife, fashion, books, and travel. Each category contains subcategories. Qloo’s knowledge of a user's taste in one category can be utilized to offer suggestions in other categories. Users then rate the suggestions, providing it with feedback for future suggestions. Qloo has partnerships with companies such as Expedia and iTunes. == Technology == Qloo’s Taste AI technology uses machine learning to decode and predict consumers’ interests, maintaining user anonymity. It is powered by 3.7 billion lifestyle entities (brands, music, film, TV, dining, nightlife, fashion, books, travel, and more) and trillions of anonymized consumer behavioral signals. Through AI, Qloo identifies patterns in these data signals, making predictions about how much interest a person or group has in a concept or thing. Central to Qloo’s technology are algorithms designed to detect and mitigate biases within datasets and models, allowing Qloo to assess the fairness of its AI systems with a focus on attributes such as age, gender, and race, enabling the company to fine-tune its AI models to align with their ethical standards. They also use visualization tools to probe the behavior of their AI models for conducting counterfactual analyses and for comparing the performances of the AI models across diverse demographic segments. Qloo’s Taste AI doesn’t collect or use any Personally Identifiable Information (PII). Instead, it derives recommendations for audience segments based on co-occurrences between lifestyle entities and anonymized behavioral signals. == Applications == Starbucks uses Qloo to create in-store music playlists tailored to specific neighborhoods. Hershey’s uses Qloo to customize the content of assorted candy bags. Michelin uses Qloo to serve recommendations in its Michelin Guide app. Netflix leverages Qloo’s technology to enhance merchandising by identifying actors who resonate with certain demographics. Qloo also works with PepsiCo, Samsung, The New York Mets, BuzzFeed, and Ticketmaster, Universal Music Group, and OOH advertising company JCDecaux.

Affectiva

Affectiva is an artificial intelligence software development company. In 2021, the company was acquired by SmartEye. The company claimed its AI understood human emotions, cognitive states, activities and the objects people use, by analyzing facial and vocal expressions. The offshoot of MIT Media Lab, Affectiva created a new technological category of artificial emotional intelligence, namely, Emotion AI. == History == Affectiva was co-founded by Rana el Kaliouby, who became chief executive officer as of May 25, 2016, and Rosalind W. Picard, who worked as chairman and Chief Scientist until 2013. Both of Affectiva's early products grew out of collaborative research at the MIT's Media Lab to help people on the autism spectrum. Affectiva was acquired for a mostly-stock deal of $73.5m by Swedish SmartEye, a former competitor. == Technology == The company has expanded its Emotion AI technology to detect more than facial expressions, reactions and emotions. Affectiva's software detects complex and nuanced emotions, cognitive states, such as drowsiness and distraction, certain activities and the objects people use. It does that by analyzing the human face, vocal intonations and body posture. Affectiva's AI is built with deep learning, computer vision, and large amounts of data that has been collected in real-world scenarios. The AI uses an optical sensor like a webcam or smartphone camera to identify a human face in real-time. Then, computer vision algorithms identify key features on the face, which are analyzed by deep learning algorithms to classify facial expressions. These facial expressions are then mapped back to emotions. One journal paper found the Affectiva iMotions Facial Expression Analysis Software results are comparable to results using facial Electromyography. Affectiva also uses computer vision to detect objects like a cellphone and car seat, as well as body key points, which track body joints to determine movement and location. Affectiva has collected massive amounts of data that are used to train and test the company's deep learning algorithms, and provide insight into human emotional reactions and engagement. The company has analyzed more than 10 million face videos from 90 countries, making it one of the largest data repositories of its kind. Affectiva has also collected more than 19,000 hours of automotive in-cabin data from 4,000 unique individuals. This automotive data is used to adapt its algorithms to varying camera angles, lighting and other environmental conditions in a vehicle. === Applications === Affectiva's AI had many applications, but the company's primary focus is on Media Analytics. Other uses of Affectiva's AI includes applications in automotive, healthcare and mental health, robotics, conversational interfaces, education, gaming, and more. ==== Media analytics ==== Affectiva's technology was first deployed in media analytics, for market research purposes. The company had since then tested more than 53,000 ads in 90 countries. Brands, advertising agencies and insights firms used the company's Emotion AI to measure the unfiltered and unbiased emotional responses consumers have when viewing video ads and movie trailers. These insights helped improve brand and media content, and predict key metrics in advertising such as sales lift, purchase intent and virality. Affectiva's technology was also used in qualitative research. Affectiva had partnered with leading insights firms such as Kantar, LRW, Added Value and Unruly. Through these collaborations, 28 percent of the Fortune Global 500 companies, and 70 percent of the world's largest advertisers, used Affectiva's Emotion AI. On September 5, 2019, Affectiva announced the appointment of Graham Page, a seasoned Kantar executive, as Global Managing Director of Media Analytics to expand on the company's existing footprint in the media analytics space. ==== Automotive ==== On March 21, 2018, Affectiva launched Affectiva Automotive AI, the first multi-modal in-cabin sensing solution to understand what is happening with people in a vehicle. It used cameras in the car to measure in real time, the state of the driver, the state of the occupants and the state of the vehicle interior (i.e. cabin). This insight helped car manufacturers, fleet management companies and rideshare providers improve road safety and build better driver monitoring systems, by understanding dangerous driver behavior such as drowsiness, distraction and anger. It was also used to create more comfortable and enjoyable transportation experiences, by understanding how passengers react to the environment, such as content they can consume in the back of the car. In addition to understanding driver and occupant emotional and cognitive states, Affectiva Automotive AI could also detect contextual cabin information such as the number of passengers, where they are sitting and if an object is present. Affectiva worked with a number of leading car manufacturers and transportation technology companies, including Aptiv, Cerence, Hyundai Kia, Faurecia, Porsche, BMW, GreenRoad Technologies, and Veoneer. == Acquisition == In June 2021 Smart Eye acquired Affectiva.