Mountain Car, a standard testing domain in Reinforcement learning, is a problem in which an under-powered car must drive up a steep hill. Since gravity is stronger than the car's engine, even at full throttle, the car cannot simply accelerate up the steep slope. The car is situated in a valley and must learn to leverage potential energy by driving up the opposite hill before the car is able to make it to the goal at the top of the rightmost hill. The domain has been used as a test bed in various reinforcement learning papers. == Introduction == The mountain car problem, although fairly simple, is commonly applied because it requires a reinforcement learning agent to learn on two continuous variables: position and velocity. For any given state (position and velocity) of the car, the agent is given the possibility of driving left, driving right, or not using the engine at all. In the standard version of the problem, the agent receives a negative reward at every time step when the goal is not reached; the agent has no information about the goal until an initial success. == History == The mountain car problem appeared first in Andrew Moore's PhD thesis (1990). It was later more strictly defined in Singh and Sutton's reinforcement learning paper with eligibility traces. The problem became more widely studied when Sutton and Barto added it to their book Reinforcement Learning: An Introduction (1998). Throughout the years many versions of the problem have been used, such as those which modify the reward function, termination condition, and the start state. == Techniques used to solve mountain car == Q-learning and similar techniques for mapping discrete states to discrete actions need to be extended to be able to deal with the continuous state space of the problem. Approaches often fall into one of two categories, state space discretization or function approximation. === Discretization === In this approach, two continuous state variables are pushed into discrete states by bucketing each continuous variable into multiple discrete states. This approach works with properly tuned parameters but a disadvantage is information gathered from one state is not used to evaluate another state. Tile coding can be used to improve discretization and involves continuous variables mapping into sets of buckets offset from one another. Each step of training has a wider impact on the value function approximation because when the offset grids are summed, the information is diffused. === Function approximation === Function approximation is another way to solve the mountain car. By choosing a set of basis functions beforehand, or by generating them as the car drives, the agent can approximate the value function at each state. Unlike the step-wise version of the value function created with discretization, function approximation can more cleanly estimate the true smooth function of the mountain car domain. === Eligibility traces === One aspect of the problem involves the delay of actual reward. The agent is not able to learn about the goal until a successful completion. Given a naive approach for each trial the car can only backup the reward of the goal slightly. This is a problem for naive discretization because each discrete state will only be backed up once, taking a larger number of episodes to learn the problem. This problem can be alleviated via the mechanism of eligibility traces, which will automatically backup the reward given to states before, dramatically increasing the speed of learning. Eligibility traces can be viewed as a bridge from temporal difference learning methods to Monte Carlo methods. == Technical details == The mountain car problem has undergone many iterations. This section focuses on the standard well-defined version from Sutton (2008). === State variables === Two-dimensional continuous state space. V e l o c i t y = ( − 0.07 , 0.07 ) {\displaystyle Velocity=(-0.07,0.07)} P o s i t i o n = ( − 1.2 , 0.6 ) {\displaystyle Position=(-1.2,0.6)} === Actions === One-dimensional discrete action space. m o t o r = ( l e f t , n e u t r a l , r i g h t ) {\displaystyle motor=(left,neutral,right)} === Reward === For every time step: r e w a r d = − 1 {\displaystyle reward=-1} === Update function === For every time step: A c t i o n = [ − 1 , 0 , 1 ] {\displaystyle Action=[-1,0,1]} V e l o c i t y = V e l o c i t y + ( A c t i o n ) ∗ 0.001 + cos ( 3 ∗ P o s i t i o n ) ∗ ( − 0.0025 ) {\displaystyle Velocity=Velocity+(Action)0.001+\cos(3Position)(-0.0025)} P o s i t i o n = P o s i t i o n + V e l o c i t y {\displaystyle Position=Position+Velocity} === Starting condition === Optionally, many implementations include randomness in both parameters to show better generalized learning. P o s i t i o n = − 0.5 {\displaystyle Position=-0.5} V e l o c i t y = 0.0 {\displaystyle Velocity=0.0} === Termination condition === End the simulation when: P o s i t i o n ≥ 0.6 {\displaystyle Position\geq 0.6} == Variations == There are many versions of the mountain car which deviate in different ways from the standard model. Variables that vary include but are not limited to changing the constants (gravity and steepness) of the problem so specific tuning for specific policies become irrelevant and altering the reward function to affect the agent's ability to learn in a different manner. An example is changing the reward to be equal to the distance from the goal, or changing the reward to zero everywhere and one at the goal. Additionally, a 3D mountain car can be used, with a 4D continuous state space.
Plotly
Plotly is a technical computing company headquartered in Montreal, Quebec, that develops online data analytics and visualization tools. Plotly provides online graphing, analytics, and statistics tools for individuals and collaboration, as well as scientific graphing libraries for Python, R, MATLAB, Perl, Julia, Arduino, JavaScript and REST. == History == Plotly was founded by Alex Johnson, Jack Parmer, Chris Parmer, and Matthew Sundquist. The founders' backgrounds are in science, energy, and data analysis and visualization. Early employees include Christophe Viau, a Canadian software engineer and Ben Postlethwaite, a Canadian geophysicist. Plotly was named one of the Top 20 Hottest Innovative Companies in Canada by the Canadian Innovation Exchange. Plotly was featured in "startup row" at PyCon 2013, and sponsored the SciPy 2018 conference. Plotly raised $5.5 million during its Series A funding, led by MHS Capital, Siemens Venture Capital, Rho Ventures, Real Ventures, and Silicon Valley Bank. The Boston Globe and Washington Post newsrooms have produced data journalism using Plotly. In 2020, Plotly was named a Best Place to Work by the Canadian SME National Business Awards, and nominated as Business of the Year. == Products == Plotly offers open-source and enterprise products. Dash is an open-source Python, R, and Julia framework for building web-based analytic applications. Many specialized open-source Dash libraries exist that are tailored for building domain-specific Dash components and applications. Some examples are Dash DAQ, for building data acquisition GUIs to use with scientific instruments, and Dash Bio, which enables users to build custom chart types, sequence analysis tools, and 3D rendering tools for bioinformatics applications. Dash Enterprise is Plotly's paid product for building, testing, deploying, managing and scaling Dash applications organization-wide. Chart Studio Cloud is a free, online tool for creating interactive graphs. It has a point-and-click graphical user interface for importing and analyzing data into a grid and using stats tools. Graphs can be embedded or downloaded. Chart Studio Enterprise is a paid product that allows teams to create, style, and share interactive graphs on a single platform. It offers expanded authentication and file export options, and does not limit sharing and viewing. Data visualization libraries Plotly.js is an open-source JavaScript library for creating graphs and powers Plotly.py for Python, as well as Plotly.R for R, MATLAB, Node.js, Julia, and Arduino and a REST API. Plotly can also be used to style interactive graphs with Jupyter notebook. Figure converters which convert matplotlib, ggplot2, and IGOR Pro graphs into interactive, online graphs. == Data visualization libraries == Plotly provides a collection of supported chart types across several programming languages: == Dash == Dash is a Python framework built on top of React, a JavaScript library. Dash also works for R, and most recently supports Julia. While still described as a Python framework, Python isn't used for the other languages: "... describing Dash as a Python framework misses a key feature of its design: the Python side (the back end/server) of Dash was built to be lightweight and stateless [allowing] multiple back-end languages to coexist on an equal footing". It is possible to integrate D3.js charts as Dash components. Dash provides the default CSS (plus HTML and JavaScript), but for custom styling Dash applications, CSS can be added, or Dash Enterprise used. === Dash Enterprise === Dash Enterprise is Plotly's paid product for building, testing, deploying, managing and scaling Dash applications organization-wide. The product integrates with enterprise IT systems to enable organizations to build, deploy and scale low-code Dash applications. With open-source Dash, analytic applications can be run from a local machine, but cannot be easily accessed by others in the organization. ==== Enterprise IT integration ==== Dash Enterprise installs on cloud environments and on-premises. Amazon Web Services, Google Cloud Platform, and Microsoft Azure are supported, as are multiple Linux on-premises servers. Authentication integrations include LDAP, AD, PKI, Okta, SAML, OAuth2, SSO, and email authentication, and Dash application access is managed through a GUI rather than code. Dash Enterprise connects to major big data backends, including Salesforce, PostgreSQL, Databricks via PySpark, Snowflake, Dask, Datashader, and Vaex. In 2020, Plotly partnered with NVIDIA to integrate Dash with RAPIDS, and NVIDIA participated in Plotly's Series C funding round. ==== Low-code capabilities ==== Dash Enterprise enables low-code development of Dash applications, which is not possible with open-source Dash. Enterprise users can write applications in multiple development environments, including Jupyter Notebook. Dash Enterprise ships with several “development engines” for drag-and-drop application editing, application design, and automated reporting, as well as dozens of artificial intelligence and machine learning application templates. ==== Deployment and scaling ==== Dash application code is deployed to Dash Enterprise using the git-push command. Dash application deployments are containerized to avoid dependency conflicts, and can be embedded in existing web platforms without iframes. Deployed applications can be managed and accessed in a single portal called App Manager, where administrators can control user authentication and view usage analytics. Dash Enterprise scales horizontally with Kubernetes. Jobs queuing, GPU acceleration, and CPU parallelization support high performance computing requirements. Plotly also offers professional services for application development and workshop training.
Simple Knowledge Organization System
Simple Knowledge Organization System (SKOS) is a W3C recommendation designed for representation of thesauri, classification schemes, taxonomies, subject-heading systems, or any other type of structured controlled vocabulary. SKOS is part of the Semantic Web family of standards built upon RDF and RDFS, and its main objective is to enable easy publication and use of such vocabularies as linked data. == History == === DESIRE II project (1997–2000) === The most direct ancestor to SKOS was the RDF Thesaurus work undertaken in the second phase of the EU DESIRE project . Motivated by the need to improve the user interface and usability of multi-service browsing and searching, a basic RDF vocabulary for Thesauri was produced. As noted later in the SWAD-Europe workplan, the DESIRE work was adopted and further developed in the SOSIG and LIMBER projects. A version of the DESIRE/SOSIG implementation was described in W3C's QL'98 workshop, motivating early work on RDF rule and query languages: A Query and Inference Service for RDF. === LIMBER (1999–2001) === SKOS built upon the output of the Language Independent Metadata Browsing of European Resources (LIMBER) project funded by the European Community, and part of the Information Society Technologies programme. In the LIMBER project CCLRC further developed an RDF thesaurus interchange format which was demonstrated on the European Language Social Science Thesaurus (ELSST) at the UK Data Archive as a multilingual version of the English language Humanities and Social Science Electronic Thesaurus (HASSET) which was planned to be used by the Council of European Social Science Data Archives CESSDA. === SWAD-Europe (2002–2004) === SKOS as a distinct initiative began in the SWAD-Europe project, bringing together partners from both DESIRE, SOSIG (ILRT) and LIMBER (CCLRC) who had worked with earlier versions of the schema. It was developed in the Thesaurus Activity Work Package, in the Semantic Web Advanced Development for Europe (SWAD-Europe) project. SWAD-Europe was funded by the European Community, and part of the Information Society Technologies programme. The project was designed to support W3C's Semantic Web Activity through research, demonstrators and outreach efforts conducted by the five project partners, ERCIM, the ILRT at Bristol University, HP Labs, CCLRC and Stilo. The first release of SKOS Core and SKOS Mapping were published at the end of 2003, along with other deliverables on RDF encoding of multilingual thesauri and thesaurus mapping. === Semantic web activity (2004–2005) === Following the termination of SWAD-Europe, SKOS effort was supported by the W3C Semantic Web Activity in the framework of the Best Practice and Deployment Working Group. During this period, focus was put both on consolidation of SKOS Core, and development of practical guidelines for porting and publishing thesauri for the Semantic Web. === Development as W3C Recommendation (2006–2009) === The SKOS main published documents — the SKOS Core Guide, the SKOS Core Vocabulary Specification, and the Quick Guide to Publishing a Thesaurus on the Semantic Web — were developed through the W3C Working Draft process. Principal editors of SKOS were Alistair Miles, initially Dan Brickley, and Sean Bechhofer. The Semantic Web Deployment Working Group, chartered for two years (May 2006 – April 2008), put in its charter to push SKOS forward on the W3C Recommendation track. The roadmap projected SKOS as a Candidate Recommendation by the end of 2007, and as a Proposed Recommendation in the first quarter of 2008. The main issues to solve were determining its precise scope of use, and its articulation with other RDF languages and standards used in libraries (such as Dublin Core). === Formal release (2009) === On August 18, 2009, W3C released the new standard that builds a bridge between the world of knowledge organization systems – including thesauri, classifications, subject headings, taxonomies, and folksonomies – and the linked data community, bringing benefits to both. Libraries, museums, newspapers, government portals, enterprises, social networking applications, and other communities that manage large collections of books, historical artifacts, news reports, business glossaries, blog entries, and other items can now use SKOS to leverage the power of linked data. === Historical view of components === SKOS was originally designed as a modular and extensible family of languages, organized as SKOS Core, SKOS Mapping, and SKOS Extensions, and a Metamodel. The entire specification is now complete within the namespace http://www.w3.org/2004/02/skos/core#. == Overview == In addition to the reference itself, the SKOS Primer (a W3C Working Group Note) summarizes the Simple Knowledge Organization System. The SKOS defines the classes and properties sufficient to represent the common features found in a standard thesaurus. It is based on a concept-centric view of the vocabulary, where primitive objects are not terms, but abstract notions represented by terms. Each SKOS concept is defined as an RDF resource. Each concept can have RDF properties attached, including: one or more preferred index terms (at most one in each natural language) alternative terms or synonyms definitions and notes, with specification of their language Concepts can be organized in hierarchies using broader-narrower relationships, or linked by non-hierarchical (associative) relationships. Concepts can be gathered in concept schemes, to provide consistent and structured sets of concepts, representing whole or part of a controlled vocabulary. === Element categories === The principal element categories of SKOS are concepts, labels, notations, documentation, semantic relations, mapping properties, and collections. The associated elements are listed in the table below. === Concepts === The SKOS vocabulary is based on concepts. Concepts are the units of thought—ideas, meanings, or objects and events (instances or categories)—which underlie many knowledge organization systems. As such, concepts exist in the mind as abstract entities which are independent of the terms used to label them. In SKOS, a Concept (based on the OWL Class) is used to represent items in a knowledge organization system (terms, ideas, meanings, etc.) or such a system's conceptual or organizational structure. A ConceptScheme is analogous to a vocabulary, thesaurus, or other way of organizing concepts. SKOS does not constrain a concept to be within a particular scheme, nor does it provide any way to declare a complete scheme—there is no way to say the scheme consists only of certain members. A topConcept is (one of) the upper concept(s) in a hierarchical scheme. === Labels and notations === Each SKOS label is a string of Unicode characters, optionally with language tags, that are associated with a concept. The prefLabel is the preferred human-readable string (maximum one per language tag), while altLabel can be used for alternative strings, and hiddenLabel can be used for strings that are useful to associate, but not meant for humans to read. A SKOS notation is similar to a label, but this literal string has a datatype, like integer, float, or date; the datatype can even be made up (see 6.5.1 Notations, Typed Literals and Datatypes in the SKOS Reference). The notation is useful for classification codes and other strings not recognizable as words. === Documentation === The Documentation or Note properties provide basic information about SKOS concepts. All the properties are considered a type of skos:note; they just provide more specific kinds of information. The property definition, for example, should contain a full description of the subject resource. More specific note types can be defined in a SKOS extension, if desired. A query for skos:note ? will obtain all the notes about , including definitions, examples, and scope, history and change, and editorial documentation. Any of these SKOS Documentation properties can refer to several object types: a literal (e.g., a string); a resource node that has its own properties; or a reference to another document, for example using a URI. This enables the documentation to have its own metadata, like creator and creation date. Specific guidance on SKOS documentation properties can be found in the SKOS Primer Documentary Notes. === Semantic relations === SKOS semantic relations are intended to provide ways to declare relationships between concepts within a concept scheme. While there are no restrictions precluding their use with two concepts from separate schemes, this is discouraged because it is likely to overstate what can be known about the two schemes, and perhaps link them inappropriately. The property related simply makes an association relationship between two concepts; no hierarchy or generality relation is implied. The properties broader and narrower are used to assert a direct hierarchical link between two concepts. The meaning may be unexpected; the relat
Jakub Pachocki
Jakub Pachocki (born 1991) is a Polish computer scientist and former competitive programmer. He is best known as OpenAI's chief scientist and for his role in overseeing development of GPT-4. == Background == Pachocki was born in 1991 in Gdańsk, Poland. In high school, he was a six-time finalist of the Polish Olympiad in Informatics. In 2009, he qualified for the International Olympiad in Informatics, winning a silver medal. Pachocki obtained his undergraduate degree in Computer Science from the University of Warsaw. He represented his university at the International Collegiate Programming Contest with his team winning a gold medal and coming second place overall in 2012. In the same year he was also the champion of the Google Code Jam. From 2011 to 2012, Pachocki worked at Facebook as a software engineering intern. Pachocki attended graduate school at Carnegie Mellon University, where he obtained his PhD under the supervision of Gary Miller. == Career == After graduation, Pachocki did postdoc work at Harvard University and Simons Institute for the Theory of Computing. === OpenAI === In 2017, Pachocki joined OpenAI. In 2021, he became OpenAI's research director where he led the development of GPT-4 and OpenAI Five. In May 2024, he became chief scientist after his mentor Ilya Sutskever left the company. OpenAI CEO Sam Altman has called Pachocki "easily one of the greatest minds of our generation". == Competitive programming achievements == International Olympiad in Informatics: Silver medal (2009) International Collegiate Programming Contest World Finals: Gold medal (second place overall in 2012) Google Code Jam: Champion (2012), Third place (2011) Facebook Hacker Cup: Second place (2013) TopCoder Open Algorithm: Second place (2012) A more comprehensive list of achievements can be found at the Competitive Programming Hall Of Fame website.
Artificial intelligence safety institute
An artificial intelligence safety institute is a type of state-backed organization aiming to evaluate and ensure the safety of advanced artificial intelligence (AI) models, also called frontier AI models. AI safety gained prominence in 2023, notably with public declarations about potential existential risks from AI. During the AI Safety Summit in November 2023, the United Kingdom and the United States both created their own AISI. During the AI Seoul Summit in May 2024, international leaders agreed to form a network of AI Safety Institutes, comprising institutes from the UK, the US, Japan, France, Germany, Italy, Singapore, South Korea, Australia, Canada and the European Union. In 2025, the UK's AI Safety Institute was renamed the "AI Security Institute", and its US counterpart became the Center for AI Standards and Innovation (CAISI). == Timeline == In 2023, Rishi Sunak, the Prime Minister of the United Kingdom, expressed his intention to "make the UK not just the intellectual home but the geographical home of global AI safety regulation" and unveiled plans for an AI Safety Summit. He emphasized the need for independent safety evaluations, stating that AI companies cannot "mark their own homework". During the summit in November 2023, the UK AISI was officially established as an evolution of the Frontier AI Taskforce, and the US AISI as part of the National Institute of Standards and Technology. Japan followed by launching an AI safety institute in February 2024. Politico reported in April 2024 that many AI companies had not shared pre-deployment access to their most advanced AI models for evaluation. Meta's president of global affairs Nick Clegg said that many AI companies were waiting for the UK and the US AI Safety Institutes to work out common evaluation rules and procedures. An agreement was indeed concluded between the UK and the US in April 2024 to collaborate on at least one joint safety test. Initially established in London, the UK AI Safety Institute announced in May 2024 that it would open an office in San Francisco, where many AI companies are located. This is part of a plan to "set new, international standards on AI safety", according to UK's technology minister Michele Donelan. == International network == At the AI Seoul Summit in May 2024, the European Union and other countries agreed to create their own AI safety institutes, forming an international network. In July 2025, the international network held an exercise to explore issues with evaluating AI agents, especially when it came to leaking sensitive information or cybersecurity. Network members also met at NeurIPS 2025 in the city of San Diego. == Specific institutes == === Australia === The Albanese government announced the creation of the Australian AI Safety Institute on 25 November 2025. === Canada === Canada announced in April 2024 that it would create an AI safety institute, and such an institute was officially founded in November 2024. The institute is housed under Innovation, Science and Economic Development Canada, though it also partners with the Canadian Institute for Advanced Research (CIFAR). It is supported by a budget of CA$50,000,000 for a five-year timespan. === European Union === The EU AI office, founded in May 2024, is a member of the international network of AI safety institutes. === France === On 31 January 2025, the government of France created the Institut national pour l'évaluation et la sécurité de l'intelligence artificielle (INESIA), or the National Institute for AI Evaluation and Security. === India === The Ministry of Electronics and Information Technology held consultations with Meta Platforms, Google, Microsoft, IBM, OpenAI, NASSCOM, Broadband India Forum, Software Alliance, Indian Institutes of Technology (IITs), The Quantum Hub, Digital Empowerment Foundation, and Access Now on October 7, 2024, in relation to the establishment of the AI Safety Institute. The decision was made to shift focus from regulation to standards-setting, risk identification, and damage detection—all of which require interoperable technologies. The AISI may spend the ₹20 crore allotted to the Safe and Trusted Pillar of the IndiaAI Mission for the initial budget. Future funding may come from other components of the IndiaAI Mission. UNESCO and MeitY began consulting on AI Readiness Assessment Methodology under Safety and Ethics in Artificial Intelligence from 2024. It is to encourage the ethical and responsible use of AI in industries. The study will find areas where government can become involved, especially in attempts to strengthen institutional and regulatory capabilities. Minister for Electronics & Information Technology Ashwini Vaishnaw announced the creation of an IndiaAI Safety Institute on January 30, 2025, to ensure the ethical and safe application of AI models. The institute will promote domestic R&D that is grounded in India's social, economic, cultural, and linguistic diversity and is based on Indian datasets. With the help of academic and research institutions, as well as private sector partners, the institute will follow the hub-and-spoke approach to carry out projects within Safe and Trusted Pillar of the IndiaAI Mission. It operates under a "hub-and-spoke" model with collaboration from academic institutions (e.g., IITs), tech firms, and international organizations like UNESCO. === Japan === The Japan AISI (or J-AISI) was founded in February 2024. Part of the Information Technology Promotion Agency, it employs about 23 people. The institute consists of the Council of AISI, the AISI Steering Committee, and a secretariat with six teams. Akiko Murakami (previously of IBM Japan and Sompo Japan) serves as the institute's executive director, and Kenji Hiramoto and Suguru Nishimura serve as the institute's two deputy executive directors. === Kenya === Kenya agreed to join the international network of AI safety institutes, but the country has not announced any details yet. It is the only African state in the network. === Singapore === The Digital Trust Centre was initially founded in June 2022. In May 2024, it was renamed to the Singapore AISI. Part of Nanyang Technological University, the institute partners with Infocomm Media Development Authority and is supported by an investment of S$10,000,000 per year. === South Korea === South Korea announced in May 2024 that it would create an AI safety institute under the umbrella of the Electronics and Telecommunications Research Institute. It will be supported by a tentative investment of somewhere between 10 and 20 million South Korean won per year, and employ at least 30 people. The institute was founded in November 2024 and is based in Bundang District within the city of Seongnam. === United Kingdom === The United Kingdom founded in April 2023 a safety organisation called Frontier AI Taskforce, with an initial budget of £100 million. In November 2023, it evolved into the AI Safety Institute, and continued to be led by Ian Hogarth. The AISI is part of the United Kingdom's Department for Science, Innovation and Technology. The United Kingdom's AI strategy aims to balance safety and innovation. Unlike the European Union which adopted the AI Act, the UK is reluctant to legislate early, considering that it may lower the sector's growth, and that laws might be rendered obsolete by technological progress. In May 2024, the institute open-sourced an AI safety tool called "Inspect", which evaluates AI model capabilities such as reasoning and their degree of autonomy. In February 2025, the UK body was renamed the AI Security Institute. Observers saw the name change as a signal that the institute will not focus on ethical issues such as algorithmic bias or freedom of speech in AI applications. === United States === The US AISI was founded in November 2023 as part of the National Institute of Standards and Technology (NIST). This happened the day after the signature of the Executive Order 14110. In February 2024, Joe Biden's former economic policy adviser Elizabeth Kelly was appointed to lead it. In February 2024, the US government created the US AI Safety Institute Consortium (AISIC), regrouping more than 200 organizations such as Google, Anthropic or Microsoft. In March 2024, a budget of $10 million was allocated. Observers noted that this investment is relatively small, especially considering the presence of many big AI companies in the US. The NIST itself, which hosts the AISI, is also known for its chronic lack of funding. Biden administration's request for additional funding was met with further budget cuts from congressional appropriators. Under President Trump, plans for members of the agency to attend the February 2025 AI Action Summit in Paris were scrapped. The US and the UK refused to sign the summit's final communique. US Vice President JD Vance said "pro-growth AI policies" should be prioritised over safety. The name of the agency was changed in June 2025 to the Center for AI Standards and Innovation
Space-based data center
Space-based data centers or orbital AI infrastructure are proposed concepts to build AI data centers in the sun-synchronous orbit or other orbits utilizing space-based solar power. Electric power has become the main bottleneck for terrestrial AI infrastructure. Space-based edge computing has historical roots in military architectures designed to bypass the latency of ground-based targeting networks. In the 1980s, the Strategic Defense Initiative's Brilliant Pebbles program first envisioned autonomous on-orbit data processing for missile defense. In 2019, the Space Development Agency (SDA) began to revive this decentralized approach through its Proliferated Warfighter Space Architecture (PWSA). This ambitious "sensor-to-shooter" infrastructure is treated as a prerequisite for the modern Golden Dome program, which would rely on space-based data processing to continuously track targets. == History == Early thinking about space-based computing infrastructure grew out of mid-20th-century visions for large orbital industrial systems, most notably proposals for space-based solar power, which were popularized in both technical literature and science writing by figures such as Isaac Asimov in the 1940s. These ideas emphasized exploiting the vacuum, continuous solar energy, and thermal characteristics of space to support power-intensive activities that would be difficult or inefficient on Earth. In the 21st century, advances in small satellites, reusable launch vehicles, and high-performance computing revived interest in space-based data centers, with governments and private companies exploring orbital or near-space platforms for edge computing, secure data handling, and low-latency processing of Earth-observation data. In September 2024, Y Combinator-backed Starcloud released a white paper detailing plans to build multiple gigawatts of AI compute in orbit. It was the first widely cited proposal to actually start building large orbital data centers. In 2025, Starcloud deployed an NVIDIA H100-class system and became the first company to train an LLM in space and run a version of Google Gemini in space. In March 2025, Lonestar deployed a data backup machine on the surface of the moon. In early January 2026, a team from the University of Pennsylvania presented a tether-based architecture for orbital data centers at the AIAA SciTech conference. The design relied on gravity gradient tension and solar-pressure-based passive attitude stabilization to minimize the mass of MW-scale orbital data centers. In January 2026, SpaceX filed plans with the Federal Communications Commission (FCC) for millions of satellites, leveraging reusable launches and Starlink integration to extend cloud and AI computing into orbit. Around the same time, Blue Origin announced the TeraWave constellation of about 5,400 satellites, designed to provide high‑throughput networking for data centers, enterprise, and government customers. Meanwhile, China announced a 200,000‑satellite constellation, focusing on state coordination, data sovereignty, and in-orbit processing for secure, time-critical applications. In February 2026, Starcloud submitted a proposal to the FCC for a constellation of up to 88,000 satellites for orbital data centers. In March, it announced intentions to be the first to mine Bitcoin in space, flying bitcoin mining ASICs on its second satellite, Starcloud-2. In May 2026, Edge Aerospace was awarded a contract by the European Space Agency under its Space Cloud program to study use cases, architectures and implementation roadmap for orbital data centers. == Feasibility == In October 2025, Nature Electronics published a study led by a research group at Nanyang Technological University on the development of carbon-neutral data centres in space. In November 2025, Google published a feasibility study on space-based data centers. The authors argued that if launch costs to low earth orbit reached US$200/kg, the launch cost for data center satellites could be cost effective relative to current energy costs for ground-based data centers. They project this may occur around 2035 if SpaceX's Starship project scales to 180 launches/year by then. == Advantages == Some sun-synchronous orbit (SSO) planes have constant sunlight in the dawn/dusk which could provide continuous solar energy. SSO is a limited resource and proper management and sharing of it is required. Solar irradiance is 36% higher in Earth orbit than on the surface No Earth weather storms or clouds, however more exposed to Solar storms. No property tax or land-use regulation. Saves space for other land use. Ample space for scalability. Won't strain the power grid. Direct access to power source without additional infrastructure. == Disadvantages == The deployment of space-based data centers raises several technical, economic, and environmental concerns. Existing launch costs are substantial and remains main cost of space infrastructure deployment Cooling is limited to heat dissipation through radiation only, which made in inefficient in comparison to convection in terrestrial data centers Space infrastructure must be designed to survive launch and to work under environment conditions of radiation, wide range of temperatures, in vacuum and in microgravity In-space assembly is on early development stage to enable deployment of mega-structures Megastructures are particularly exposed to orbital debris Solar arrays efficiency decrease 0.5% to 0.8% per year due to exposure of ultraviolet rays, space weather and orbital thermal cycles Hardware is designed for limited lifespan. Maintenance and repair in space (known as On-Orbit Servicing (OOS)) is still on early stage of practical implementation. Disposable data centre: technology obsolescence of AI data centre being a concern and difficult maintenance in space imply the single-use purpose of those space data centres. To extend lifetime, space infrastructure will require either refueling or orbit rasie by the servicer, which is going to increase its operational costs The environmental impact on Earth has its own challenges: The environmental impact of launches need to be addressed. Deployment consumes Earth resources that cannot be recovered or recycled. Computers require lots of resources, some of which are strategic. Recycling e-waste is already a challenge on Earth and extremely unlikely in space. Space debris (orbit pollution) is another sustainability challenge for space: Orbits are, like any resources, a limited physical and electromagnetic resource and available for all mankind. The accumulation of satellites on a particular orbit reduces the use of space for other purposes. A consequence of the increase of satellite in orbit is a higher risk of the runaway of space debris (see Kessler syndrome). This means some orbits could become unusable. Latency and bandwidth are constrained in space, and consumes limited electromagnetic resources. Satellite flares could inhibit ground-based and space-based observational astronomy. == Size and power generated == It would take ~1 square mile solar array in earth orbit to produce 1 gigawatt of power at 30% cell efficiency. == Companies pursuing space-based AI infrastructure == Blue Origin Cowboy Space Corporation (formerly Aetherflux) Edge Aerospace Google – Project Suncatcher Nvidia OpenAI SpaceX Starcloud
Alexey Chervonenkis
Alexey Yakovlevich Chervonenkis (Russian: Алексей Яковлевич Червоненкис; 7 September 1938 – 22 September 2014) was a Soviet and Russian mathematician. Along with Vladimir Vapnik, he was one of the main developers of the Vapnik–Chervonenkis theory, also known as the "fundamental theory of learning", an important part of computational learning theory. Chervonenkis held joint appointments with the Russian Academy of Sciences and Royal Holloway, University of London. Alexey Chervonenkis got lost in Losiny Ostrov National Park on 22 September 2014, and later during a search operation was found dead near Mytishchi, a suburb of Moscow. He had died of hypothermia.