Catastrophic interference, also known as catastrophic forgetting, is the tendency of an artificial neural network to abruptly and drastically forget previously learned information upon learning new information. Neural networks are an important part of the connectionist approach to cognitive science. The issue of catastrophic interference when modeling human memory with connectionist models was originally brought to the attention of the scientific community by research from McCloskey and Cohen (1989), and Ratcliff (1990). It is a radical manifestation of the 'sensitivity-stability' dilemma or the 'stability-plasticity' dilemma. Specifically, these problems refer to the challenge of making an artificial neural network that is sensitive to, but not disrupted by, new information. Lookup tables and connectionist networks lie on the opposite sides of the stability plasticity spectrum. The former remains completely stable in the presence of new information but lacks the ability to generalize, i.e. infer general principles, from new inputs. On the other hand, connectionist networks like the standard backpropagation network can generalize to unseen inputs, but they are sensitive to new information. Backpropagation models can be analogized to human memory insofar as they have a similar ability to generalize, but these networks often exhibit less stability than human memory. Notably, these backpropagation networks are susceptible to catastrophic interference. This is an issue when modelling human memory, because unlike these networks, humans typically do not show catastrophic forgetting. == Discovery == The term catastrophic interference was originally coined by McCloskey and Cohen (1989) but was also brought to the attention of the scientific community by research from Ratcliff (1990). === The Sequential Learning Problem: McCloskey and Cohen (1989) === McCloskey and Cohen (1989) noted the problem of catastrophic interference during two different experiments with backpropagation neural network modelling. Experiment 1: Learning the ones and twos addition facts In their first experiment they trained a standard backpropagation neural network on a single training set consisting of 17 single-digit ones problems (i.e., 1 + 1 through 9 + 1, and 1 + 2 through 1 + 9) until the network could represent and respond properly to all of them. The error between the actual output and the desired output steadily declined across training sessions, which reflected that the network learned to represent the target outputs better across trials. Next, they trained the network on a single training set consisting of 17 single-digit twos problems (i.e., 2 + 1 through 2 + 9, and 1 + 2 through 9 + 2) until the network could represent, respond properly to all of them. They noted that their procedure was similar to how a child would learn their addition facts. Following each learning trial on the twos facts, the network was tested for its knowledge on both the ones and twos addition facts. Like the ones facts, the twos facts were readily learned by the network. However, McCloskey and Cohen noted the network was no longer able to properly answer the ones addition problems even after one learning trial of the twos addition problems. The output pattern produced in response to the ones facts often resembled an output pattern for an incorrect number more closely than the output pattern for a correct number. This is considered to be a drastic amount of error. Furthermore, the problems 2+1 and 1+2, which were included in both training sets, even showed dramatic disruption during the first learning trials of the twos facts. Experiment 2: Replication of Barnes and Underwood (1959) study In their second connectionist model, McCloskey and Cohen attempted to replicate the study on retroactive interference in humans by Barnes and Underwood (1959). They trained the model on A-B and A-C lists and used a context pattern in the input vector (input pattern), to differentiate between the lists. Specifically the network was trained to respond with the right B response when shown the A stimulus and A-B context pattern and to respond with the correct C response when shown the A stimulus and the A-C context pattern. When the model was trained concurrently on the A-B and A-C items then the network readily learned all of the associations correctly. In sequential training the A-B list was trained first, followed by the A-C list. After each presentation of the A-C list, performance was measured for both the A-B and A-C lists. They found that the amount of training on the A-C list in Barnes and Underwood study that lead to 50% correct responses, lead to nearly 0% correct responses by the backpropagation network. Furthermore, they found that the network tended to show responses that looked like the C response pattern when the network was prompted to give the B response pattern. This indicated that the A-C list apparently had overwritten the A-B list. This could be likened to learning the word dog, followed by learning the word stool and then finding that you think of the word stool when presented with the word dog. McCloskey and Cohen tried to reduce interference through a number of manipulations including changing the number of hidden units, changing the value of the learning rate parameter, overtraining on the A-B list, freezing certain connection weights, changing target values 0 and 1 instead 0.1 and 0.9. However, none of these manipulations satisfactorily reduced the catastrophic interference exhibited by the networks. Overall, McCloskey and Cohen (1989) concluded that: at least some interference will occur whenever new learning alters the weights involved in representing old learning the greater the amount of new learning, the greater the disruption in old knowledge interference was catastrophic in the backpropagation networks when learning was sequential but not concurrent === Constraints Imposed by Learning and Forgetting Functions: Ratcliff (1990) === Ratcliff (1990) used multiple sets of backpropagation models applied to standard recognition memory procedures, in which the items were sequentially learned. After inspecting the recognition performance models he found two major problems: Well-learned information was catastrophically forgotten as new information was learned in both small and large backpropagation networks. Even one learning trial with new information resulted in a significant loss of the old information, paralleling the findings of McCloskey and Cohen (1989). Ratcliff also found that the resulting outputs were often a blend of the previous input and the new input. In larger networks, items learned in groups (e.g. AB then CD) were more resistant to forgetting than were items learned singly (e.g. A then B then C...). However, the forgetting for items learned in groups was still large. Adding new hidden units to the network did not reduce interference. Discrimination between the studied items and previously unseen items decreased as the network learned more. This finding contradicts studies on human memory, which indicated that discrimination increases with learning. Ratcliff attempted to alleviate this problem by adding 'response nodes' that would selectively respond to old and new inputs. However, this method did not work as these response nodes would become active for all inputs. A model which used a context pattern also failed to increase discrimination between new and old items. == Proposed solutions == The main cause of catastrophic interference seems to be overlap in the representations at the hidden layer of distributed neural networks. In a distributed representation, each input tends to create changes in the weights of many of the nodes. Catastrophic forgetting occurs because when many of the weights where "knowledge is stored" are changed, it is unlikely for prior knowledge to be kept intact. During sequential learning, the inputs become mixed, with the new inputs being superimposed on top of the old ones. Another way to conceptualize this is by visualizing learning as a movement through a weight space. This weight space can be likened to a spatial representation of all of the possible combinations of weights that the network could possess. When a network first learns to represent a set of patterns, it finds a point in the weight space that allows it to recognize all of those patterns. However, when the network then learns a new set of patterns, it will move to a place in the weight space for which the only concern is the recognition of the new patterns. To recognize both sets of patterns, the network must find a place in the weight space suitable for recognizing both the new and the old patterns. Below are a number of techniques which have empirical support in successfully reducing catastrophic interference in backpropagation neural networks: === Orthogonality === Many of the early techniques in reducing representational overlap involved making either the input vecto
Halite AI Programming Competition
Halite is an open-source computer programming contest developed by the hedge fund/tech firm Two Sigma in partnership with a team at Cornell Tech. Programmers can see the game environment and learn everything they need to know about the game. Participants are asked to build bots in whichever language they choose to compete on a two-dimensional virtual battle field. == History == Benjamin Spector and Michael Truell created the first Halite competition in 2016, before partnering with Two Sigma later that year. === Halite I === Halite I asked participants to conquer territory on a grid. It launched in November 2016 and ended in February 2017. Halite I attracted about 1,500 players. === Halite II === Halite II was similar to Halite I, but with a space-war theme. It ran from October 2017 until January 2018. The second installment of the competition attracted about 6,000 individual players from more than 100 countries. Among the participants were professors, physicists and NASA engineers, as well as high school and university students. === Halite III === Halite III launched in mid-October 2018. It ran from October 2018 to January 2019, with an ocean themed playing field. Players were asked to collect and manage Halite, an energy resource. By the end of the competition, Halite III included more than 4000 players and 460 organizations. === Halite IV === Halite IV was hosted by Kaggle, and launched in mid-June 2020.
John M. Jumper
John Michael Jumper (born 1 January 1985) is an American chemist and computer scientist. Jumper and Demis Hassabis were awarded the 2024 Nobel Prize in Chemistry for protein structure prediction. As of 2025 Jumper serves as director at Google DeepMind. Jumper and his colleagues created AlphaFold, an artificial intelligence (AI) model to predict protein structures from their amino acid sequence with high accuracy. The AlphaFold team had released 214 million protein structures as of January 2024. The scientific journal Nature included Jumper as one of the ten "people who mattered" in science in their annual listing of Nature's 10 in 2021. == Education == Jumper graduated from Pulaski Academy in 2003. He received a Bachelor of Science with majors in physics and mathematics from Vanderbilt University in 2007, a Master of Philosophy in theoretical condensed matter physics from the University of Cambridge where he was a student of St Edmund's College, Cambridge in 2010 on a Marshall Scholarship, a Master of Science in theoretical chemistry from the University of Chicago in 2012, and a Doctor of Philosophy in theoretical chemistry from the University of Chicago in 2017. His doctoral advisors at the University of Chicago were Tobin R. Sosnick and Karl Freed. == Career and research == Jumper's research investigates algorithms for protein structure prediction. === AlphaFold === AlphaFold is a deep learning algorithm developed by Jumper and his team at DeepMind, a research lab acquired by Google's parent company Alphabet Inc. It is an artificial intelligence program which performs predictions of protein structure. === Awards and honors === In November 2020, AlphaFold was named the winner of the 14th Critical Assessment of Structure Prediction (CASP) competition. This international competition benchmarks algorithms to determine which one can best predict the 3D structure of proteins. AlphaFold won the competition, outperforming other algorithms scoring above 90 for around two-thirds of the proteins in CASP's global distance test (GDT), a test that measures the degree to which a computational program predicted structure is similar to the lab experiment determined structure, with 100 being a complete match, within the distance cutoff used for calculating GDT. In 2021, Jumper was awarded the BBVA Foundation Frontiers of Knowledge Award in the category "Biology and Biomedicine". In 2022 Jumper received the Wiley Prize in Biomedical Sciences and for 2023 the Breakthrough Prize in Life Sciences for developing AlphaFold, which accurately predicts the structure of a protein. In 2023 he was awarded the Canada Gairdner International Award and the Albert Lasker Award for Basic Medical Research. In 2024, Jumper and Demis Hassabis shared half of the Nobel Prize in Chemistry for their protein folding predictions, the other half went to David Baker for computational protein design. In 2025, Jumper received the Golden Plate Award of the American Academy of Achievement and the Marshall Medal of the Marshall Aid Commemoration Commission. He was elected a Fellow of the Royal Society (FRS) that same year. In 2026, he was elected a member of the National Academy of Engineering.
Lighthill report
Artificial Intelligence: A General Survey, commonly known as the Lighthill report, is a scholarly article by James Lighthill, published in Artificial Intelligence: a paper symposium in 1973. It was compiled by Lighthill for the British Science Research Council as an evaluation of academic research in the field of artificial intelligence (AI). The report gave a very pessimistic prognosis for many core aspects of research in this field, stating that "In no part of the field have the discoveries made so far produced the major impact that was then promised". It "formed the basis for the decision by the British government to end support for AI research in most British universities", contributing to an AI winter in the United Kingdom. == Publication history == It was commissioned by the SRC in 1972 for Lighthill to "make a personal review of the subject [of AI]". Lighthill completed the report in July. The SRC discussed the report in September, and decided to publish it, together with some alternative points of view by Stuart Sutherland, Roger Needham, Christopher Longuet-Higgins, and Donald Michie. The SRC's decision to invite the report was partly a reaction to high levels of discord within the University of Edinburgh's Department of Artificial Intelligence, one of the earliest and biggest centres for AI research in the UK. On May 9, 1973, Lighthill debated several leading AI researchers (Donald Michie, John McCarthy, Richard Gregory) at the Royal Institution in London concerning the report. == Content == While the report was supportive of research into the simulation of neurophysiological and psychological processes, it was "highly critical of basic research in foundational areas such as robotics and language processing". The report stated that AI researchers had failed to address the issue of combinatorial explosion when solving problems within real-world domains. That is, the report states that whilst AI techniques may have worked within the scope of small problem domains, the techniques would not scale up well to solve more realistic problems. The report represents a pessimistic view of AI that began after early excitement in the field. The report divides AI research into three categories: Advanced Automation ("A"): applications of AI, such as optical character recognition, mechanical component design and manufacture, missile perception and guidance, etc. Computer-based Central Nervous System research ("C"): building computational models of human brains (neurobiology) and behavior (psychology). Bridge, or Building Robots ("B"): research that combines categories A and C. This category is intentionally vague. Projects in category A had had some success, but only in restricted domains where a large quantity of detailed knowledge was used in designing the program. This was disappointing to researchers who hoped for generic methods. Due to the issue of the combinatorial explosion, the amount of detailed knowledge required by the program quickly grew too large to be entered by hand, thus restricting projects to restricted domains. Projects in category C had had some measure of success. Artificial neural networks were successfully used to model neurobiological data. SHRDLU demonstrated that human use of language, even in fine details, depends on the semantics or knowledge, and is not purely syntactical. This was influential in psycholinguistics. Attempts to extend SHRDLU to larger domains of discourse was considered impractical, again due to the issue of the combinatorial explosion. Projects in category B were held to be failures. One important project, that of "programming and building a robot that would mimic human ability in a combination of eye-hand co-ordination and common-sense problem solving", was considered entirely disappointing. Similarly, chess playing programs were no better than human amateurs. Due to the combinatorial explosion, the run-time of general algorithms quickly grew impractical, requiring detailed problem-specific heuristics. The report stated that it was expected that within the next 25 years, category A would simply become applied technologies engineering, C would integrate with psychology and neurobiology, while category B would be abandoned.
ChessMachine
The ChessMachine was a chess computer sold between 1991 and 1995 by TASC (The Advanced Software Company). It was unique at the time for incorporating both an ARM2 coprocessor for the chess engine on an ISA card which plugged into an IBM PC and a software interface running on the PC to display a chess board and control the engine. The ISA card was sold with a CPU running at either 16 MHz or 32 MHz, and 128 KB, 512 KB, or 1 MB of onboard memory for transposition tables. This made economic sense at the time of introduction because mainstream PCs were only running from 10 MHz to 25 MHz. Two engines were sold with the card: The King by Johann de Koning and Gideon by Ed Schröder. Gideon was famed for winning two World Computer Chess Championships on this hardware. The King later became the engine used in the popular Chessmaster series of chess programs. TASC later incorporated the technology into a dedicated unit, sold from 1993 to 1997. There were two models, the R30 and R40, running at 30 MHz and 40 MHz respectively, and having 512 KB and 1 MB of transposition tables, respectively. The SmartBoard, a wooden sensory board, was connected to the units, which were in tiny boxes approximately the size of chess clocks. They were only sold with The King chess engine. This was the end of the era of strong dedicated chess computers, and these two models are acknowledged as the strongest dedicated chess computers that were ever sold. At the height of its strength, the R30 attained a rating over 2350 on computer rating lists, higher than any other dedicated unit. According to the SSDF rating list, the R30 held its own against its contemporary programs running a Pentium-90 MHz and won against other dedicated units.
Visual Expert
Visual Expert is a static code analysis tool, extracting design and technical information from software source code by reverse-engineering, used by programmers for software maintenance, modernization or optimization. It is designed to parse several programming languages at the same time (PL/SQL, Transact-SQL, PowerBuilder...) and analyze cross-language dependencies, in addition to each language's source code. Visual Expert checks source code against hundreds of code inspection rules for vulnerability assessment, bug fix, and maintenance issues. == Features == Cross-references exploration: Impact Analysis, E/R diagrams, call graphs, CRUD matrix, dependency graphs. Software documentation: a documentation generator produces technical documentation and low-level design descriptions. Inspect the code to detect bugs, security vulnerabilities and maintainability issues. Native integration with Jenkins. Reports on duplicate code, unused objects and methods and naming conventions. Calculates software metrics and source lines of code. Code comparison: finds differences between several versions of the same code. Performance analysis: identifies code parts that slow down the application because of their syntax - it extracts statistics about code execution from the database and combines it with the static analysis of the code. == Usage == Visual Expert is used in several contexts: Change impact analysis: evaluating the consequences of a change in the code or in a database. Avoiding negative side effects when evolving a system. Static Application Security Testing (SAST): detecting and removing security issues. Continuous Integration / Continuous Inspection : adding a static code analysis job in a CI/CD workflow to automatically verify the quality and security of a new build when it is released. Program comprehension: helping programmers understand and maintain existing code, or modernize legacy systems. Transferring knowledge of the code, from one programmer to another. Software sizing: calculating the size of an application, or a piece of code, in order to estimate development efforts. Code review: improving the code by finding and removing code smells, dead code, code causing poor performances or violations of coding conventions. == Limitations == As a static code analyzer, Visual Expert is limited to the programming languages supported by its code parsers - Oracle PL/SQL, SQL Server Transact-SQL, PowerBuilder. A preliminary reverse engineering is required. Visual Expert does it automatically, but its duration depends on the size of the code parsed. Users must wait for the parsing completion prior to using the features, or schedule it in advance. They must also allocate sufficient hardware resources to support their volume of code. Visual Expert is based on a client/server architecture: the code analysis is running on a Windows PC - preferably a server. The information extracted from the code is stored in a RDBMS, communicating with a client application installed on the programmer's computer - no web client is available. This requires that the code, the parsers, the RDBMS and the programmers’ computers are connected to the same LAN or VPN. == History == 1995- 1998 - Prog and Doc - Initial version distributed on the French market 2001 - Visual Expert 4.5 2003 - Visual Expert 5 2007 - Visual Expert 5.7 2010 - Visual Expert 6.0 2015 - Visual Expert 2015 - Server component added to schedule code analyses 2016 - Visual Expert 2016 - Oracle PL/SQL code parser, code inventory (lines of code, number of objects…) 2017 - Visual Expert 2017 - SQL Server T-SQL code parser, Code comparison, CRUD matrix 2018 - Visual Expert 2018 - DB Code Performance Analysis, integration with TFS 2019 - Visual Expert 2019 - Generation of E/R diagrams from the code 2020 - Visual Expert 2020 - Object dependency matrix, naming consistency verification, integration with GIT and SVN 2021 - Visual Expert 2021 - Continuous Code Inspection, integration with Jenkins 2022 - Visual Expert 2022 - Support for cloud-based repositories and large volumes of code 2023 - Visual Expert 2023 - Performance tuning for PowerBuilder 2024 - Visual Expert 2024 - New web UI to simplify deployment and use among large teams. 2025 - Visual Expert 2025 - AI-based features to explain code, generate comments, and optimize queries
OpenVINO
OpenVINO is an open-source software toolkit developed by Intel for optimizing and deploying deep learning models. It supports several popular model formats and categories, such as large language models, computer vision, and generative AI. OpenVINO is optimized for Intel hardware, but offers support for ARM/ARM64 processors. It sees great use in AI Sound Processing drivers when tied with Intel's Gaussian & Neural Accelerator (GNA). Based in C++, it extends API support for C and Python, as well as Node.js (in early preview). OpenVINO is cross-platform and free for use under Apache License 2.0. == Workflow == The simplest OpenVINO usage involves obtaining a model and running it as is. Yet for the best results, a more complete workflow is suggested: obtain a model in one of supported frameworks, convert the model to OpenVINO IR using the OpenVINO Converter tool, optimize the model, using training-time or post-training options provided by OpenVINO's NNCF. execute inference, using OpenVINO Runtime by specifying one of several inference modes. == OpenVINO model format == OpenVINO IR is the default format used to run inference. It is saved as a set of two files, .bin and .xml, containing weights and topology, respectively. It is obtained by converting a model from one of the supported frameworks, using the application's API or a dedicated converter. Models of the supported formats may also be used for inference directly, without prior conversion to OpenVINO IR. Such an approach is more convenient but offers fewer optimization options and lower performance, since the conversion is performed automatically before inference. Some pre-converted models can be found in the Hugging Face repository. The supported model formats are: PyTorch TensorFlow TensorFlow Lite ONNX (including formats that may be serialized to ONNX) PaddlePaddle JAX/Flax == OS support == OpenVINO runs on Windows, Linux and MacOS.