Parchive (a portmanteau of parity archive, and formally known as Parity Volume Set Specification) is an erasure code system that produces par files for checksum verification of data integrity, with the capability to perform data recovery operations that can repair or regenerate corrupted or missing data. Parchive was originally written to solve the problem of reliable file sharing on Usenet, but it can be used for protecting any kind of data from data corruption, disc rot, bit rot, and accidental or malicious damage. Despite the name, Parchive uses more advanced techniques (specifically error correction codes) than simplistic parity methods of error detection. As of 2015, PAR1 is obsolete, PAR2 is mature for widespread use, and PAR3 is a discontinued experimental version developed by MultiPar author Yutaka Sawada. The original SourceForge Parchive project has been inactive since April 30, 2015. A new PAR3 specification has been worked on since April 28, 2019 by PAR2 specification author Michael Nahas. An alpha version of the PAR3 specification has been published on January 29, 2022 while the program itself is being developed. == History == Parchive was intended to increase the reliability of transferring files via Usenet newsgroups. Usenet was originally designed for informal conversations, and the underlying protocol, NNTP was not designed to transmit arbitrary binary data. Another limitation, which was acceptable for conversations but not for files, was that messages were normally fairly short in length and limited to 7-bit ASCII text. Various techniques were devised to send files over Usenet, such as uuencoding and Base64. Later Usenet software allowed 8 bit Extended ASCII, which permitted new techniques like yEnc. Large files were broken up to reduce the effect of a corrupted download, but the unreliable nature of Usenet remained. With the introduction of Parchive, parity files could be created that were then uploaded along with the original data files. If any of the data files were damaged or lost while being propagated between Usenet servers, users could download parity files and use them to reconstruct the damaged or missing files. Parchive included the construction of small index files (.par in version 1 and .par2 in version 2) that do not contain any recovery data. These indexes contain file hashes that can be used to quickly identify the target files and verify their integrity. Because the index files were so small, they minimized the amount of extra data that had to be downloaded from Usenet to verify that the data files were all present and undamaged, or to determine how many parity volumes were required to repair any damage or reconstruct any missing files. They were most useful in version 1 where the parity volumes were much larger than the short index files. These larger parity volumes contain the actual recovery data along with a duplicate copy of the information in the index files (which allows them to be used on their own to verify the integrity of the data files if there is no small index file available). In July 2001, Tobias Rieper and Stefan Wehlus proposed the Parity Volume Set specification, and with the assistance of other project members, version 1.0 of the specification was published in October 2001. Par1 used Reed–Solomon error correction to create new recovery files. Any of the recovery files can be used to rebuild a missing file from an incomplete download. Version 1 became widely used on Usenet, but it did suffer some limitations: It was restricted to handle at most 255 files. The recovery files had to be the size of the largest input file, so it did not work well when the input files were of various sizes. (This limited its usefulness when not paired with the proprietary RAR compression tool.) The recovery algorithm had a bug, due to a flaw in the academic paper on which it was based. It was strongly tied to Usenet and it was felt that a more general tool might have a wider audience. In January 2002, Howard Fukada proposed that a new Par2 specification should be devised with the significant changes that data verification and repair should work on blocks of data rather than whole files, and that the algorithm should switch to using 16 bit numbers rather than the 8 bit numbers that PAR1 used. Michael Nahas and Peter Clements took up these ideas in July 2002, with additional input from Paul Nettle and Ryan Gallagher (who both wrote Par1 clients). Version 2.0 of the Parchive specification was published by Michael Nahas in September 2002. Peter Clements then went on to write the first two Par2 implementations, QuickPar and par2cmdline. Abandoned since 2004, Paul Houle created phpar2 to supersede par2cmdline. Yutaka Sawada created MultiPar to supersede QuickPar. MultiPar uses par2j.exe (which is partially based on par2cmdline's optimization techniques) to use as MultiPar's backend engine. == Versions == Versions 1 and 2 of the file format are incompatible. (However, many clients support both.) === Par1 === For Par1, the files f1, f2, ..., fn, the Parchive consists of an index file (f.par), which is CRC type file with no recovery blocks, and a number of "parity volumes" (f.p01, f.p02, etc.). Given all of the original files except for one (for example, f2), it is possible to create the missing f2 given all of the other original files and any one of the parity volumes. Alternatively, it is possible to recreate two missing files from any two of the parity volumes and so forth. Par1 supports up to a total of 256 source and recovery files. === Par2 === Par2 files generally use this naming/extension system: filename.vol000+01.PAR2, filename.vol001+02.PAR2, filename.vol003+04.PAR2, filename.vol007+06.PAR2, etc. The number after the "+" in the filename indicates how many blocks it contains, and the number after "vol" indicates the number of the first recovery block within the PAR2 file. If an index file of a download states that 4 blocks are missing, the easiest way to repair the files would be by downloading filename.vol003+04.PAR2. However, due to the redundancy, filename.vol007+06.PAR2 is also acceptable. There is also an index file filename.PAR2, it is identical in function to the small index file used in PAR1. Par2 specification supports up to 32,768 source blocks and up to 65,535 recovery blocks. Input files are split into multiple equal-sized blocks so that recovery files do not need to be the size of the largest input file. Although Unicode is mentioned in the PAR2 specification as an option, most PAR2 implementations do not support Unicode. Directory support is included in the PAR2 specification, but most or all implementations do not support it. === Par3 === The Par3 specification was originally planned to be published as an enhancement over the Par2 specification. However, to date, it has remained closed source by specification owner Yutaka Sawada. A discussion on a new format started in the GitHub issue section of the maintained fork par2cmdline on January 29, 2019. The discussion led to a new format which is also named as Par3. The new Par3 format's specification is published on GitHub, but remains being an alpha draft as of January 28, 2022. The specification is written by Michael Nahas, the author of Par2 specification, with the help from Yutaka Sawada, animetosho and malaire. The new format claims to have multiple advantages over the Par2 format, including support for: More than 216 files and more than 216 blocks. Packing small files into one block, as well as deduplication when a block appears in multiple files. UTF-8 file names. File permissions, hard links, symbolic/soft links, and empty directories. Embedding PAR data inside other formats, like ZIP archives or ISO disk images. "Incremental backups", where a user creates recovery files for some file or folder, change some data, and create new recovery files reusing some of the older files. More error correction code algorithms (such as LDPC and sparse random matrix). BLAKE3 hashes, dropping support for the MD5 hashes used in PAR2. == Software == === Multi-platform === par2+tbb (GPLv2) — a concurrent (multithreaded) version of par2cmdline 0.4 using TBB. Only compatible with x86 based CPUs. It is available in the FreeBSD Ports system as par2cmdline-tbb. Original par2cmdline — (obsolete). Available in the FreeBSD Ports system as par2cmdline. par2cmdline maintained fork by BlackIkeEagle. par2cmdline-mt is another multithreaded version of par2cmdline using OpenMP, GPLv2, or later. Currently merged into BlackIkeEagle's fork and maintained there. ParPar (CC0) is a high performance, multithreaded PAR2 client and Node.js library. Does not support verifying or repair, it can currently only create PAR2 archives. par2deep (LGPL-3.0) — Produce, verify and repair par2 files recursively, both on the command line as well as with the aid of a graphical user interface. It is available in the Python Package Index system as par2deep. par2cron (MIT License) is an o
Figure AI
Figure AI, Inc. is an American robotics company developing humanoid robots that operate via artificial intelligence. The company was founded in 2022 by Brett Adcock. As of late 2025, the company has a $39 billion valuation. Three generations of humanoid robots (Figure 01–03) have been developed, as well as two iterations of a vision-language-action model (Helix 01–02), which can control up to two robots at once. By 2026, the robots demonstrated the potential ability to perform household work and the company gained publicity when a Figure 03 appeared at a White House event. == History == Figure AI was founded in 2022 by Brett Adcock, also known for founding Archer Aviation and Vettery. That year, the company introduced its prototype, Figure 01, a bipedal robot designed for manual labor, initially targeting the logistics and warehousing sectors. The initial model utilized external cabling for easier maintenance. In May 2023, Figure AI raised $70 million from investors including Adcock, who invested $20 million, and Parkway Venture Capital. In January 2024, Figure AI announced a partnership with BMW to deploy humanoid robots in automotive manufacturing facilities. In February 2024, Figure AI secured $675 million in venture capital funding from a consortium that includes Jeff Bezos, Microsoft, Nvidia, Intel, and the startup-funding divisions of Amazon and OpenAI; the company was then valued at $2.6 billion. Figure AI also announced a partnership with OpenAI, which would build specialized artificial intelligence (AI) models for Figure AI's humanoid robots, enabling its robots to process language; the collaboration ended after a year, with Adcock stating that large language models had become a smaller problem compared to those allowing for "high rate robot control". In August 2024, the company introduced Figure 02, describing it as the next step toward deploying humanoids for industrial use. The machine has 35 degrees of freedom (DOF), while the five-fingered hands have 16 DOF and the ability to carry up to 25 kilograms (55 lb). The model is equipped with cabling integrated into the limbs, a torso-placed battery, six RGB cameras, and an onboard vision-language-action (VLA) model. It has three times the computing power (including inference AI) of the previous model, including two graphics processing units, supported by Nvidia. Microphones, speakers, and custom AI models (developed with OpenAI) enable communication with humans. In early 2025, Figure AI announced BotQ, a manufacturing facility aiming to produce 12,000 humanoids per year with the help of its own humanoid robots, and Helix, a VLA model that can control up to two robots at once. Helix enables a robot to interact with the world without extensive manual training, according to the company allowing it to pick up nearly any small household object. By April, the company issued cease-and-desist letters to at least two secondary brokers promoting its private stock without authorization. In September, a third round of financing exceeded $1 billion, raising the company's total valuation to $39 billion. Investors included Brookfield Asset Management, Intel, Macquarie Capital, Nvidia, Parkway Venture Capital, Qualcomm, Salesforce, and T-Mobile. In October 2025, Figure 03 was introduced. According to the company, its hardware and software redesign aims to create a general-purpose robot able to learn directly from humans. An upgraded camera system delivers twice the frame rate, a quarter the latency, and a 60% wider field of view, in addition to a camera in each hand. Tactile sensors in the fingertips can detect forces as little as 3 grams (0.1 oz). It incorporates soft materials and a protected battery for safety, and removable, washable textiles. It supports wireless inductive charging. In November 2025, the former head of product safety sued the company on the basis of being fired for raising the concern that the company's robots were strong enough to fracture a human skull. By early 2026, Figure 02 had been used in demonstrations showing that it could load a washing machine, sort packages, and fold laundry. That January, Helix 02 was released, expanding the AI model to the entire body to allow for functional autonomy. A Helix 02–powered Figure 02 was shown to be capable of loading and unloading a dishwasher, based on hours of motion-capture data and simulation-based machine learning. In March, U.S. First Lady Melania Trump appeared at the White House with a Figure 03, promoting the presumptive eventual ability of AI to teach children. In May 2026, Figure AI livestreamed a group of their robots processing packages nonstop for almost a week, inspiring a 10-hour competition between their robot and a human, in which the robot performed 98.5% as well as the human.
Ontology for Biomedical Investigations
The Ontology for Biomedical Investigations (OBI) is an open-access, integrated ontology for the description of biological and clinical investigations. OBI provides a model for the design of an investigation, the protocols and instrumentation used, the materials used, the data generated and the type of analysis performed on it. The project is being developed as part of the OBO Foundry and as such adheres to all the principles therein such as orthogonal coverage (i.e. clear delineation from other foundry member ontologies) and the use of a common formal language. In OBI the common formal language used is the Web Ontology Language (OWL). As of March 2008, a pre-release version of the ontology was made available at the project's SVN repository. == Scope == The Ontology for Biomedical Investigations (OBI) addresses the need for controlled vocabularies to support integration and joint ("cross-omics") analysis of experimental data, a need originally identified in the transcriptomics domain by the FGED Society, which developed the MGED Ontology as an annotation resource for microarray data.Smith B, Ashburner M, Rosse C, Bard J, Bug W, Ceusters W, et al. (November 2007). "The OBO Foundry: coordinated evolution of ontologies to support biomedical data integration". Nature Biotechnology. 25 (11): 1251–5. doi:10.1038/nbt1346. PMC 2814061. PMID 17989687. OBI uses the basic formal ontology upper-level ontology as a means of describing general entities that do not belong to a specific problem domain. As such, all OBI classes are a subclass of some BFO class. The ontology has the scope of modeling all biomedical investigations and as such contains ontology terms for aspects such as: biological material – for example blood plasma instrument (and parts of an instrument therein) – for example DNA microarray, centrifuge information content – such as an image or a digital information entity such as an electronic medical record design and execution of an investigation (and individual experiments therein) – for example study design, electrophoresis material separation data transformation (incorporating aspects such as data normalization and data analysis) – for example principal components analysis dimensionality reduction, mean calculation Less 'concrete' aspects such as the role a given entity may play in a particular scenario (for example the role of a chemical compound in an experiment) and the function of an entity (for example the digestive function of the stomach to nutriate the body) are also covered in the ontology. == OBI consortium == The MGED Ontology was originally identified in the transcriptomics domain by the FGED Society and was developed to address the needs of data integration. Following a mutual decision to collaborate, this effort later became a wider collaboration between groups such as FGED, PSI and MSI in response to the needs of areas such as transcriptomics, proteomics and metabolomics and the FuGO (Functional Genomics Investigation Ontology) was created. This later became the OBI covering the wider scope of all biomedical investigations. As an international, cross-domain initiative, the OBI consortium draws upon a pool of experts from a variety of fields, not limited to biology. The current list of OBI consortium members is available at the OBI consortium website. The consortium is made up of a coordinating committee which is a combination of two subgroups, the Community Representative (those representing a particular biomedical community) and the Core Developers (ontology developers who may or may not be members of any single community). Separate to the coordinating committee is the Developers Working Group which consists of developers within the communities collaborating in the development of OBI at the discretion of current OBI Consortium members. == Papers on OBI ==
Microsoft Office PerformancePoint Server
Microsoft Office PerformancePoint Server is a business intelligence software product released in 2007 by Microsoft. The product was generally an integration of the acquisitions from ProClarity - the Planning Server and Monitoring Server - into Microsoft's SharePoint server product line. Although discontinued in 2009, the dashboard, scorecard, and analytics capabilities of PerformancePoint Server were incorporated into SharePoint 2010 and later versions. PerformancePoint Server also provided a planning and budgeting component directly integrated with Excel. == History == Microsoft offered preview releases of PerformancePoint Server starting in mid-2006. Previews of the product were formed from Business Scorecard Manager 2005 and the Planning Server component. Acquisitions ProClarity and Great Plains brought additional analytics and planning/reporting capabilities, as well as companion products ProClarity 6.3 and FRx. PerformancePoint Server was officially released in November 2007. Microsoft discontinued PerformancePoint Server as an independent product in 2009 and folded its dashboard, scorecard and analytics capabilities into PerformancePoint Services in SharePoint Server 2010. == Monitoring Server Component == Business monitoring capabilities, including dashboards, scorecards & key performance indicators, navigable reports for deeper analysis, strategy maps, and linked filtering, are provided by PerformancePoint's Monitoring Server component. A Dashboard Designer application that is distributed from Monitoring Server enables business analysts or IT Administrators to: create & test data source connections create views that use those data connections assemble the views into a dashboard deploy the dashboard as a SharePoint page Dashboard Designer saved content and security information back to the Monitoring Server. Data source connections, such as OLAP cubes or relational tables, were also made through Monitoring Server. After a dashboard has been published to the Monitoring Server database, it would be deployed as a SharePoint page and shared with other users as such. When the pages were opened in a web browser, Monitoring Server updated the data in the views by connecting back to the original data sources. == Planning Server Component == PerformancePoint's Planning Server component supported maintenance of logical business models, budget & approval workflows, enterprise data sources, and it followed Generally Accepted Accounting Principles. Planning Server made use of Excel for input and line-of-business reporting, as well as SQL Server for storing and processing business models. == Management Reporter Component == The Management Reporter component was designed to perform financial reporting and can read PerformancePoint Planning models directly. A development kit was also available to allow this component to read other models.
Artificial intelligence industry in Taiwan
The artificial intelligence (AI) industry in Taiwan refers to the development, application, and commercialization of artificial intelligence technologies within Taiwan. The industry has grown alongside Taiwan's established strengths in semiconductor manufacturing and information and communications technology (ICT), and is supported by government policy, research institutions, and private sector participation. AI development in Taiwan has focused on integrating hardware capabilities with software applications across sectors such as manufacturing, healthcare, and smart infrastructure. Artificial intelligence has been identified as a strategic area of development in Taiwan since the late 2010s. While Taiwan has historically played a limited role in early theoretical and expert-system phases of AI development, its position in global electronics manufacturing has provided a foundation for participation in the contemporary era of machine learning and data-driven AI systems. Taiwan's AI industry is characterized by a strong hardware base, particularly in semiconductor production and AI server manufacturing, combined with increasing investment in software, data infrastructure, and applied AI services. The sector has been shaped by global demand for computing power, advances in deep learning, and the expansion of AI applications in industrial and commercial contexts. == Government policy and development == The Taiwanese government has promoted AI development through a series of national strategies. In 2017, the Ministry of Science and Technology launched the "AI Grand Strategy for a Small Country" initiative, investing approximately US$517 million between 2017 and 2021 to support research, infrastructure, and talent development. This initiative aimed to build a domestic AI ecosystem by funding research centers, expanding data infrastructure, and supporting industrial adoption. The Executive Yuan also introduced the AI Taiwan Action Plan 1.0 (2018–2021), which focused on integrating AI technologies into existing industries and strengthening research and development capabilities. A subsequent plan, AI Taiwan Action Plan 2.0 (2023–2026), expanded the focus to include ethical governance, regulatory frameworks, and risk management in response to the growth of generative AI technologies. In 2023, the Taiwan AI Center of Excellence (Taiwan AICoE), a government-backed hub, was established by the National Science and Technology Council to accelerate AI development, foster international collaboration, and train talent in Taiwan. It acts as a specialized think tank focusing on creating a "smart technology island" by integrating AI resources and developing trusted, human-centric AI technologies. In 2024, the Taiwan Chip-based Industrial Innovation Program (CbI) was launched by the Executive Yuan as a 10-year, NT$300 billion (US$9.3 billion) initiative to leverage Taiwan's semiconductor dominance, driving innovation in AI, smart mobility, manufacturing, and healthcare. It aims to combine generative AI with IC technology, cultivate talent, and attract global startups to build a "Silicon Island". In parallel, the Taiwanese government has explored legislative frameworks such as a proposed Artificial Intelligence Fundamental Act in December 2025, addressing issues including data protection, safety standards, and intellectual property. == Industrial structure == === Semiconductor and hardware foundation === Taiwan's AI industry is closely linked to its semiconductor sector. In 2020, Taiwan accounted for approximately 77.3% of the global wafer foundry market and 57.7% of packaging and testing, with a 20.1% share in integrated circuit (IC) design. These capabilities provide critical infrastructure for AI systems, which rely on high-performance computing hardware. Taiwanese firms are also involved in the production of AI servers and related components, contributing significantly to global supply chains for data centers and cloud computing. The integration of chip design, manufacturing, and assembly has enabled Taiwan to play a central role in providing the computational resources required for AI development. On 20 November 2025, Google established the "Google Taiwan AI Infrastructure R&D Center", second only to its US headquarters and largest AI hardware infrastructure engineering center outside of the United States. === Software and services === Compared to its hardware capabilities, Taiwan's AI software sector is less developed. The absence of large-scale global AI platform companies has been noted as a structural limitation. As a result, much of Taiwan's AI industry focuses on applied solutions, including customization of existing AI models for specific industries. Therefore, efforts to strengthen software capabilities have included investment in research institutions, startup ecosystems, and collaborations between academia and industry. == Applications == === Smart manufacturing === AI has been widely applied in Taiwan's manufacturing sector, which is a major component of the economy. Applications include process automation, predictive maintenance, quality control, and fault detection. AI-enabled smart manufacturing systems aim to improve efficiency, reduce production costs, and enhance product quality. Taiwan's manufacturing industry has incorporated AI technologies into production lines, particularly in electronics and machinery sectors. === Healthcare === The use of AI in healthcare in Taiwan has expanded in areas such as medical imaging, diagnostics, and drug development. AI systems are used to analyze CT scans, MRI data, and other clinical information to support diagnosis and treatment planning. Taiwan's healthcare sector, which includes medical devices, pharmaceuticals, and medical services, has benefited from the integration of AI technologies, particularly in precision medicine and clinical decision support systems. A notable example of AI healthcare deployment in Taiwan is the collaboration between Siemens Healthineers, Ever Fortune AI, and Asia University Hospital. === Edge computing and IoT === AI applications in Taiwan increasingly involve edge computing, where data processing occurs near the source rather than in centralized cloud systems. This approach reduces latency and bandwidth requirements and is used in smart devices, sensors, and industrial equipment. Edge AI technologies are applied in areas such as smart appliances, industrial automation, and transportation systems. == Education and talent development == Human capital development has been a key focus of Taiwan's AI strategy. The Taiwan AI Academy, established in 2018 with support from Academia Sinica and industry partners, provides training programs for professionals and students aimed at accelerating the adoption of artificial intelligence technologies across industries. The academy offers a range of courses, including executive-level programs, technical training, and specialized tracks in areas such as smart manufacturing, smart healthcare, and edge AI. These programs are designed to provide intensive and practical instruction over relatively short periods. A notable component of the curriculum is project-based learning, in which participants are required to complete proof-of-concept (POC) projects addressing real-world industrial problems. These projects are often developed further for implementation within companies, facilitating technology transfer and commercialization. Between 2018 and 2021, more than 8,000 individuals completed AI training programs across campuses in Taipei, Hsinchu, Taichung, and Tainan. Graduates of the academy have contributed to the introduction of AI systems in sectors such as manufacturing, healthcare, and finance, supporting broader industrial transformation efforts. In addition to the Taiwan AI Academy, universities and research institutions in Taiwan play a significant role in AI education and research. Leading universities have expanded programs in computer science, data science, and machine learning, while research institutes conduct applied and fundamental studies in artificial intelligence. Collaboration between academia, government, and industry is a common feature of Taiwan's AI ecosystem, with joint research projects, internship programs, and technology incubation initiatives supporting talent development. Government-supported initiatives have also sought to attract and retain AI talent, including funding for graduate education, international collaboration programs, and incentives for industry–academic partnerships. These efforts aim to address talent shortages and strengthen Taiwan's capacity in both applied and foundational AI research. == Regulation and governance == Taiwan has developed guidelines and policy frameworks to address the risks associated with AI technologies. In 2023, the Executive Yuan issued guidelines for the use of generative AI in government agencies, focusing on data security and privacy. Ongoing policy discussions hav
Energy-based model
An energy-based model (EBM), also called Canonical Ensemble Learning (CEL) or Learning via Canonical Ensemble (LCE), is an application of canonical ensemble formulation from statistical physics for learning from data. The approach prominently appears in generative artificial intelligence. EBMs provide a unified framework for many probabilistic and non-probabilistic approaches to such learning, particularly for training graphical and other structured models. An EBM learns the characteristics of a target dataset and generates a similar but larger dataset. EBMs detect the latent variables of a dataset and generate new datasets with a similar distribution. Energy-based generative neural networks is a class of generative models, which aim to learn explicit probability distributions of data in the form of energy-based models, the energy functions of which are parameterized by modern deep neural networks. Boltzmann machines are a special form of energy-based models with a specific parametrization of the energy. == Description == For a given input x {\displaystyle x} , the model describes an energy E θ ( x ) {\displaystyle E_{\theta }(x)} such that the Boltzmann distribution P θ ( x ) = e − β E θ ( x ) Z ( θ ) {\displaystyle P_{\theta }(x)={e^{-\beta E_{\theta }(x)} \over Z(\theta )}} is a probability (density), and typically β = 1 {\displaystyle \beta =1} . Since the normalization constant: Z ( θ ) := ∫ x ∈ X e − β E θ ( x ) d x {\displaystyle Z(\theta ):=\int _{x\in X}e^{-\beta E_{\theta }(x)}dx} (also known as the partition function) depends on all the Boltzmann factors of all possible inputs x {\displaystyle x} , it cannot be easily computed or reliably estimated during training simply using standard maximum likelihood estimation. However, for maximizing the likelihood during training, the gradient of the log-likelihood of a single training example x {\displaystyle x} is given by using the chain rule: ∂ θ log ( P θ ( x ) ) = E x ′ ∼ P θ [ ∂ θ E θ ( x ′ ) ] − ∂ θ E θ ( x ) ( ∗ ) {\displaystyle \partial _{\theta }\log \left(P_{\theta }(x)\right)=\mathbb {E} _{x'\sim P_{\theta }}[\partial _{\theta }E_{\theta }(x')]-\partial _{\theta }E_{\theta }(x)\,()} The expectation in the above formula for the gradient can be approximately estimated by drawing samples x ′ {\displaystyle x'} from the distribution P θ {\displaystyle P_{\theta }} using Markov chain Monte Carlo (MCMC). Early energy-based models, such as the 2003 Boltzmann machine by Hinton, estimated this expectation via blocked Gibbs sampling. Newer approaches make use of more efficient Stochastic Gradient Langevin Dynamics (LD), drawing samples using: x 0 ′ ∼ P 0 , x i + 1 ′ = x i ′ − α 2 ∂ E θ ( x i ′ ) ∂ x i ′ + ϵ {\displaystyle x_{0}'\sim P_{0},x_{i+1}'=x_{i}'-{\frac {\alpha }{2}}{\frac {\partial E_{\theta }(x_{i}')}{\partial x_{i}'}}+\epsilon } , where ϵ ∼ N ( 0 , α ) {\displaystyle \epsilon \sim {\mathcal {N}}(0,\alpha )} . A replay buffer of past values x i ′ {\displaystyle x_{i}'} is used with LD to initialize the optimization module. The parameters θ {\displaystyle \theta } of the neural network are therefore trained in a generative manner via MCMC-based maximum likelihood estimation: the learning process follows an "analysis by synthesis" scheme, where within each learning iteration, the algorithm samples the synthesized examples from the current model by a gradient-based MCMC method (e.g., Langevin dynamics or Hybrid Monte Carlo), and then updates the parameters θ {\displaystyle \theta } based on the difference between the training examples and the synthesized ones – see equation ( ∗ ) {\displaystyle ()} . This process can be interpreted as an alternating mode seeking and mode shifting process, and also has an adversarial interpretation. Essentially, the model learns a function E θ {\displaystyle E_{\theta }} that associates low energies to correct values, and higher energies to incorrect values. After training, given a converged energy model E θ {\displaystyle E_{\theta }} , the Metropolis–Hastings algorithm can be used to draw new samples. The acceptance probability is given by: P a c c ( x i → x ∗ ) = min ( 1 , P θ ( x ∗ ) P θ ( x i ) ) . {\displaystyle P_{acc}(x_{i}\to x^{})=\min \left(1,{\frac {P_{\theta }(x^{})}{P_{\theta }(x_{i})}}\right).} == History == The term "energy-based models" was first coined in a 2003 JMLR paper where the authors defined a generalisation of independent components analysis to the overcomplete setting using EBMs. Other early work on EBMs proposed models that represented energy as a composition of latent and observable variables. == Characteristics == EBMs demonstrate useful properties: Simplicity and stability. The EBM is the only object that needs to be designed and trained. Separate networks need not be trained to ensure balance. Adaptive computation time. An EBM can generate sharp, diverse samples or (more quickly) coarse, less diverse samples. Given infinite time, this procedure produces true samples. Flexibility. In Variational Autoencoders (VAE) and flow-based models, the generator learns a map from a continuous space to a (possibly) discontinuous space containing different data modes. EBMs can learn to assign low energies to disjoint regions (multiple modes). Adaptive generation. EBM generators are implicitly defined by the probability distribution, and automatically adapt as the distribution changes (without training), allowing EBMs to address domains where generator training is impractical, as well as minimizing mode collapse and avoiding spurious modes from out-of-distribution samples. Compositionality. Individual models are unnormalized probability distributions, allowing models to be combined through product of experts or other hierarchical techniques. == Experimental results == On image datasets such as CIFAR-10 and ImageNet 32x32, an EBM model generated high-quality images relatively quickly. It supported combining features learned from one type of image for generating other types of images. It was able to generalize using out-of-distribution datasets, outperforming flow-based and autoregressive models. EBM was relatively resistant to adversarial perturbations, behaving better than models explicitly trained against them with training for classification. == Applications == Target applications include natural language processing, robotics and computer vision. The first energy-based generative neural network is the generative ConvNet proposed in 2016 for image patterns, where the neural network is a convolutional neural network. The model has been generalized to various domains to learn distributions of videos, and 3D voxels. They are made more effective in their variants. They have proven useful for data generation (e.g., image synthesis, video synthesis, 3D shape synthesis, etc.), data recovery (e.g., recovering videos with missing pixels or image frames, 3D super-resolution, etc), data reconstruction (e.g., image reconstruction and linear interpolation ). == Alternatives == EBMs compete with techniques such as variational autoencoders (VAEs), generative adversarial networks (GANs) or normalizing flows. == Extensions == === Joint energy-based models === Joint energy-based models (JEM), proposed in 2020 by Grathwohl et al., allow any classifier with softmax output to be interpreted as energy-based model. The key observation is that such a classifier is trained to predict the conditional probability p θ ( y | x ) = e f → θ ( x ) [ y ] ∑ j = 1 K e f → θ ( x ) [ j ] for y = 1 , … , K and f → θ = ( f 1 , … , f K ) ∈ R K , {\displaystyle p_{\theta }(y|x)={\frac {e^{{\vec {f}}_{\theta }(x)[y]}}{\sum _{j=1}^{K}e^{{\vec {f}}_{\theta }(x)[j]}}}\ \ {\text{ for }}y=1,\dotsc ,K{\text{ and }}{\vec {f}}_{\theta }=(f_{1},\dotsc ,f_{K})\in \mathbb {R} ^{K},} where f → θ ( x ) [ y ] {\displaystyle {\vec {f}}_{\theta }(x)[y]} is the y-th index of the logits f → {\displaystyle {\vec {f}}} corresponding to class y. Without any change to the logits it was proposed to reinterpret the logits to describe a joint probability density: p θ ( y , x ) = e f → θ ( x ) [ y ] Z ( θ ) , {\displaystyle p_{\theta }(y,x)={\frac {e^{{\vec {f}}_{\theta }(x)[y]}}{Z(\theta )}},} with unknown partition function Z ( θ ) {\displaystyle Z(\theta )} and energy E θ ( x , y ) = − f θ ( x ) [ y ] {\displaystyle E_{\theta }(x,y)=-f_{\theta }(x)[y]} . By marginalization, we obtain the unnormalized density p θ ( x ) = ∑ y p θ ( y , x ) = ∑ y e f → θ ( x ) [ y ] Z ( θ ) =: e − E θ ( x ) , {\displaystyle p_{\theta }(x)=\sum _{y}p_{\theta }(y,x)=\sum _{y}{\frac {e^{{\vec {f}}_{\theta }(x)[y]}}{Z(\theta )}}=:e^{-E_{\theta }(x)},} therefore, E θ ( x ) = − log ( ∑ y e f → θ ( x ) [ y ] Z ( θ ) ) , {\displaystyle E_{\theta }(x)=-\log \left(\sum _{y}{\frac {e^{{\vec {f}}_{\theta }(x)[y]}}{Z(\theta )}}\right),} so that any classifier can be used to define an energy function E θ ( x ) {\displaystyle E_{\theta }(x)} .
Affective computing
Affective computing is the study and development of systems and devices that can recognize, interpret, process, and simulate human affects. It is an interdisciplinary field spanning computer science, psychology, and cognitive science. While some core ideas in the field may be traced as far back as to early philosophical inquiries into emotion, the modern idea originated with Rosalind Picard's 1995 paper entitled "Affective Computing" and her 1997 book of the same name published by MIT Press. One motivation for researching affective computing is the ability to give machines emotional intelligence, including simulating empathy. The goal is that a machine should interpret the emotional state of humans and adapt its behavior to those emotions, responding appropriately. Recent experimental research has shown that subtle affective haptic feedback can shape human reward learning and mobile interaction behavior, suggesting that affective computing systems may not only interpret emotional states but also actively modulate user actions through emotion-laden outputs. == Areas == === Detecting and recognizing emotional information === Detecting emotional information usually begins with passive sensors that capture data about the user's physical state or behavior without interpreting the input. The data gathered is analogous to the cues humans use to perceive emotions in others. For example, a video camera might capture facial expressions, body posture, and gestures, while a microphone might capture speech. Other sensors detect emotional cues by directly measuring physiological data, such as skin temperature and galvanic resistance. Recognizing emotional information requires the extraction of meaningful patterns from the gathered data. This is done using machine learning techniques that process different modalities, such as speech recognition, natural language processing, or facial expression detection. The goal of most of these techniques is to produce labels that would match the labels a human would give in the same situation. For example, if a person makes a facial expression furrowing their brow, then the computer vision system might be trained to label their face as appearing "confused" or as "concentrating" or "slightly negative" (as opposed to positive, which it might say if they were smiling in a happy-appearing way). This response is based on the data used to train the system. These labels may or may not correspond to what the person is actually feeling. === Emotion in machines === Another area within affective computing is the design of computational devices proposed to exhibit either innate emotional capabilities or that are capable of convincingly simulating emotions. A more practical approach, based on current technological capabilities, is the simulation of emotions in conversational agents in order to enrich and facilitate interactivity between human and machine. Marvin Minsky, one of the pioneering computer scientists in artificial intelligence, relates emotions to the broader issues of machine intelligence stating in The Emotion Machine that emotion is "not especially different from the processes that we call 'thinking.'" The innovative approach "digital humans" or virtual humans includes an attempt to give these programs, which simulate humans, an emotional dimension as well, including reactions, facial expressions, and gestures in accordance with the reaction that a real person would have in certain emotionally stimulating situations. Emotion in machines often refers to emotion in computational, often AI-based, systems. As a result, the terms 'emotional AI' is being used. Some modern large language models simulate emotions in their chats with humans. ChatGPT's simulated emotion leans more positive than that of most human responses. == Technologies == In psychology, cognitive science, and in neuroscience, there have been two main approaches for describing how humans perceive and classify emotion: continuous or categorical. The continuous approach tends to use dimensions such as negative vs. positive, calm vs. aroused. The categorical approach tends to use discrete classes such as happy, sad, angry, fearful, surprise, and disgust. Different kinds of machine learning regression and classification models are used for machines to produce continuous or discrete labels. Sometimes, models are also built that allow combinations across the categories (e.g. a happy-surprised face or a fearful-surprised face). The following sections consider many of the kinds of input data used for the task of emotion recognition. === Emotional speech === Various changes in the autonomic nervous system can indirectly alter a person's speech, and affective technologies can leverage this information to recognize emotion. For example, speech produced in a state of fear, anger, or joy becomes fast, loud, and precisely enunciated, with a higher and wider range in pitch, whereas emotions such as tiredness, boredom, or sadness tend to generate slow, low-pitched, and slurred speech. Some emotions have been found to be more easily computationally identified, such as anger or approval. Emotional speech processing technologies recognize the user's emotional state using computational analysis of speech features. Vocal parameters and prosodic features such as pitch variables and speech rate can be analyzed through pattern recognition techniques. Speech analysis is an effective method of identifying affective state, having an average reported accuracy of 70-80% in research from 2003 and 2006. These systems tend to outperform average human accuracy (approximately 60%) but are less accurate than systems which employ other modalities for emotion detection, such as physiological states or facial expressions. However, since many speech characteristics are independent of semantics or culture, this technique is considered to be a promising route for further research. ==== Algorithms ==== The process of speech/text affect detection requires the creation of a reliable database, knowledge base, or vector space model, broad enough to fit every need for its application, as well as the selection of a successful classifier which will allow for quick and accurate emotion identification. As of 2010, the most frequently used classifiers were linear discriminant classifiers (LDC), k-nearest neighbor (k-NN), Gaussian mixture model (GMM), support vector machines (SVM), artificial neural networks (ANN), decision tree algorithms, and hidden Markov models (HMMs). Various studies showed that choosing the appropriate classifier can significantly enhance the overall performance of the system. The list below gives a brief description of each algorithm: LDC – Classification happens based on the value obtained from the linear combination of the feature values, which are usually provided in the form of vector features. k-NN – Classification happens by locating the object in the feature space, and comparing it with the k nearest neighbors (training examples). The majority vote decides on the classification. GMM – A probabilistic model used for representing the existence of subpopulations within the overall population. Each sub-population is described using the mixture distribution, which allows for classification of observations into the sub-populations. SVM – A type of (usually binary) linear classifier which decides in which of the two (or more) possible classes, each input may fall into. ANN – is a mathematical model, inspired by biological neural networks, that can better grasp possible non-linearities of the feature space. Decision tree algorithms – work based on following a decision tree in which leaves represent the classification outcome, and branches represent the conjunction of subsequent features that lead to the classification. HMMs – a statistical Markov model in which the states and state transitions are not directly available to observation. Instead, the series of outputs dependent on the states are visible. In the case of affect recognition, the outputs represent the sequence of speech feature vectors, which allow the deduction of states' sequences through which the model progressed. The states can consist of various intermediate steps in the expression of an emotion, and each of them has a probability distribution over the possible output vectors. The states' sequences allow us to predict the affective state which we are trying to classify, and this is one of the most commonly used techniques within the area of speech affect detection. It has been proven that having enough acoustic evidence available the emotional state of a person can be classified by a set of majority voting classifiers. The proposed set of classifiers is based on three main classifiers: kNN, C4.5 and SVM-RBF Kernel. This set achieves better performance than each basic classifier taken separately. It is compared with two other sets of classifiers: one-against-all (OAA) multiclass SVM with Hybrid kernels and th