In computer science, constrained clustering is a class of semi-supervised learning algorithms. Typically, constrained clustering incorporates either a set of must-link constraints, cannot-link constraints, or both, with a data clustering algorithm. A cluster in which the members conform to all must-link and cannot-link constraints is called a chunklet. == Types of constraints == Both a must-link and a cannot-link constraint define a relationship between two data instances. Together, the sets of these constraints act as a guide for which a constrained clustering algorithm will attempt to find chunklets (clusters in the dataset which satisfy the specified constraints). A must-link constraint is used to specify that the two instances in the must-link relation should be associated with the same cluster. A cannot-link constraint is used to specify that the two instances in the cannot-link relation should not be associated with the same cluster. Some constrained clustering algorithms will abort if no such clustering exists which satisfies the specified constraints. Others will try to minimize the amount of constraint violation should it be impossible to find a clustering which satisfies the constraints. Constraints could also be used to guide the selection of a clustering model among several possible solutions. == Examples == Examples of constrained clustering algorithms include: COP K-means PCKmeans (Pairwise Constrained K-means) CMWK-Means (Constrained Minkowski Weighted K-Means)
Neural scaling law
In machine learning, a neural scaling law is an empirical scaling law that describes how neural network performance changes as key factors are scaled up or down. These factors typically include the number of parameters, training dataset size, and training cost. Some models also exhibit performance gains by scaling inference through increased test-time compute (TTC), extending neural scaling laws beyond training to the deployment phase. == Introduction == In general, a deep learning model can be characterized by four parameters: model size, training dataset size, training cost, and the post-training error rate (e.g., the test set error rate). Each of these variables can be defined as a real number, usually written as N , D , C , L {\displaystyle N,D,C,L} (respectively: parameter count, dataset size, computing cost, and loss). A neural scaling law is a theoretical or empirical statistical law between these parameters. There are also other parameters with other scaling laws. === Size of the model === In most cases, the model's size is simply the number of parameters. However, one complication arises with the use of sparse models, such as mixture-of-expert models. With sparse models, during inference, only a fraction of their parameters are used. In comparison, most other kinds of neural networks, such as transformer models, always use all their parameters during inference. === Size of the training dataset === The size of the training dataset is usually quantified by the number of data points within it. Larger training datasets are typically preferred, as they provide a richer and more diverse source of information from which the model can learn. This can lead to improved generalization performance when the model is applied to new, unseen data. However, increasing the size of the training dataset also increases the computational resources and time required for model training. With the "pretrain, then finetune" method used for most large language models, there are two kinds of training dataset: the pretraining dataset and the finetuning dataset. Their sizes have different effects on model performance. Generally, the finetuning dataset is less than 1% the size of pretraining dataset. In some cases, a small amount of high quality data suffices for finetuning, and more data does not necessarily improve performance. Many scaling laws, due to their inherent diminishing returns nature, value data based on a submodular set function which was shown in a paper on this topic. === Cost of training === Training cost is typically measured in terms of time (how long it takes to train the model) and computational resources (how much processing power and memory are required). It is important to note that the cost of training can be significantly reduced with efficient training algorithms, optimized software libraries, and parallel computing on specialized hardware such as GPUs or TPUs. The cost of training a neural network model is a function of several factors, including model size, training dataset size, the training algorithm complexity, and the computational resources available. In particular, doubling the training dataset size does not necessarily double the cost of training, because one may train the model for several times over the same dataset (each being an "epoch"). === Performance === The performance of a neural network model is evaluated based on its ability to accurately predict the output given some input data. Common metrics for evaluating model performance include: Negative log-likelihood per token (logarithm of perplexity) for language modeling; Accuracy, precision, recall, and F1 score for classification tasks; Mean squared error (MSE) or mean absolute error (MAE) for regression tasks; Elo rating in a competition against other models, such as gameplay or preference by a human judge. Performance can be improved by using more data, larger models, different training algorithms, regularizing the model to prevent overfitting, and early stopping using a validation set. When the performance is a number bounded within the range of [ 0 , 1 ] {\displaystyle [0,1]} , such as accuracy, precision, etc., it often scales as a sigmoid function of cost, as seen in the figures. == Examples == === (Hestness, Narang, et al, 2017) === The 2017 paper is a common reference point for neural scaling laws fitted by statistical analysis on experimental data. Previous works before the 2000s, as cited in the paper, were either theoretical or orders of magnitude smaller in scale. Whereas previous works generally found the scaling exponent to scale like L ∝ D − α {\displaystyle L\propto D^{-\alpha }} , with α ∈ { 0.5 , 1 , 2 } {\displaystyle \alpha \in \{0.5,1,2\}} , the paper found that α ∈ [ 0.07 , 0.35 ] {\displaystyle \alpha \in [0.07,0.35]} . Of the factors they varied, only task can change the exponent α {\displaystyle \alpha } . Changing the architecture optimizers, regularizers, and loss functions, would only change the proportionality factor, not the exponent. For example, for the same task, one architecture might have L = 1000 D − 0.3 {\displaystyle L=1000D^{-0.3}} while another might have L = 500 D − 0.3 {\displaystyle L=500D^{-0.3}} . They also found that for a given architecture, the number of parameters necessary to reach lowest levels of loss, given a fixed dataset size, grows like N ∝ D β {\displaystyle N\propto D^{\beta }} for another exponent β {\displaystyle \beta } . They studied machine translation with LSTM ( α ∼ 0.13 {\displaystyle \alpha \sim 0.13} ), generative language modelling with LSTM ( α ∈ [ 0.06 , 0.09 ] , β ≈ 0.7 {\displaystyle \alpha \in [0.06,0.09],\beta \approx 0.7} ), ImageNet classification with ResNet ( α ∈ [ 0.3 , 0.5 ] , β ≈ 0.6 {\displaystyle \alpha \in [0.3,0.5],\beta \approx 0.6} ), and speech recognition with two hybrid (LSTMs complemented by either CNNs or an attention decoder) architectures ( α ≈ 0.3 {\displaystyle \alpha \approx 0.3} ). === (Henighan, Kaplan, et al, 2020) === A 2020 analysis studied statistical relations between C , N , D , L {\displaystyle C,N,D,L} over a wide range of values and found similar scaling laws, over the range of N ∈ [ 10 3 , 10 9 ] {\displaystyle N\in [10^{3},10^{9}]} , C ∈ [ 10 12 , 10 21 ] {\displaystyle C\in [10^{12},10^{21}]} , and over multiple modalities (text, video, image, text to image, etc.). In particular, the scaling laws it found are (Table 1 of ): For each modality, they fixed one of the two C , N {\displaystyle C,N} , and varying the other one ( D {\displaystyle D} is varied along using D = C / 6 N {\displaystyle D=C/6N} ), the achievable test loss satisfies L = L 0 + ( x 0 x ) α {\displaystyle L=L_{0}+\left({\frac {x_{0}}{x}}\right)^{\alpha }} where x {\displaystyle x} is the varied variable, and L 0 , x 0 , α {\displaystyle L_{0},x_{0},\alpha } are parameters to be found by statistical fitting. The parameter α {\displaystyle \alpha } is the most important one. When N {\displaystyle N} is the varied variable, α {\displaystyle \alpha } ranges from 0.037 {\displaystyle 0.037} to 0.24 {\displaystyle 0.24} depending on the model modality. This corresponds to the α = 0.34 {\displaystyle \alpha =0.34} from the Chinchilla scaling paper. When C {\displaystyle C} is the varied variable, α {\displaystyle \alpha } ranges from 0.048 {\displaystyle 0.048} to 0.19 {\displaystyle 0.19} depending on the model modality. This corresponds to the β = 0.28 {\displaystyle \beta =0.28} from the Chinchilla scaling paper. Given fixed computing budget, optimal model parameter count is consistently around N o p t ( C ) = ( C 5 × 10 − 12 petaFLOP-day ) 0.7 = 9.0 × 10 − 7 C 0.7 {\displaystyle N_{opt}(C)=\left({\frac {C}{5\times 10^{-12}{\text{petaFLOP-day}}}}\right)^{0.7}=9.0\times 10^{-7}C^{0.7}} The parameter 9.0 × 10 − 7 {\displaystyle 9.0\times 10^{-7}} varies by a factor of up to 10 for different modalities. The exponent parameter 0.7 {\displaystyle 0.7} varies from 0.64 {\displaystyle 0.64} to 0.75 {\displaystyle 0.75} for different modalities. This exponent corresponds to the ≈ 0.5 {\displaystyle \approx 0.5} from the Chinchilla scaling paper. It's "strongly suggested" (but not statistically checked) that D o p t ( C ) ∝ N o p t ( C ) 0.4 ∝ C 0.28 {\displaystyle D_{opt}(C)\propto N_{opt}(C)^{0.4}\propto C^{0.28}} . This exponent corresponds to the ≈ 0.5 {\displaystyle \approx 0.5} from the Chinchilla scaling paper. The scaling law of L = L 0 + ( C 0 / C ) 0.048 {\displaystyle L=L_{0}+(C_{0}/C)^{0.048}} was confirmed during the training of GPT-3 (Figure 3.1 ). === Chinchilla scaling (Hoffmann, et al, 2022) === One particular scaling law ("Chinchilla scaling") states that, for a large language model (LLM) autoregressively trained for one epoch, with a cosine learning rate schedule, we have: { C = C 0 N D L = A N α + B D β + L 0 {\displaystyle {\begin{cases}C=C_{0}ND\\L={\frac {A}{N^{\alpha }}}+{\frac {B}{D^{\beta }}}+L_{0}\end{cases}}} where the variables are C {\displaystyle C} is the cost o
AAAI Conference on Artificial Intelligence
The AAAI Conference on Artificial Intelligence is a leading international academic conference in artificial intelligence held annually. It ranks 4th in terms of H5 Index in Google Scholar's list of top AI publications, after ICLR, NeurIPS, and ICML. It is supported by the Association for the Advancement of Artificial Intelligence (AAAI), after which it is named. Precise dates vary from year to year, but paper submissions are generally due at the end of August to beginning of September, and the conference is generally held during the following February. The first AAAI was held in 1980 at Stanford University, Stanford California. During AAAI-20 conference, AI pioneers and 2018 Turing Award winners (often referred to as the Nobel Prize of Computing) Yann LeCun and Yoshua Bengio, among eight other researchers, were honored as the AAAI 2020 Fellows. Along with other conferences such as NeurIPS and ICML, AAAI uses an artificial-intelligence algorithm to assign papers to reviewers. == Sponsors == Many leading technology companies, including Google, Microsoft, Amazon (company), IBM, Baidu, Bytedance, and Huawei, generously sponsor and participate in AAAI to publish and showcase their latest theoretical and applied research. Sponsoring companies also actively recruit AI talents at the conference. == Locations == AAAI-2026 Singapore Expo, Singapore AAAI-2025 Pennsylvania Convention Center, Philadelphia, Pennsylvania, United States AAAI-2024 Vancouver Convention Centre, Vancouver, British Columbia, Canada AAAI-2023 Washington Convention Center, Washington, D.C., United States AAAI-2022 Virtual Conference AAAI-2021 Virtual Conference AAAI-2020 Hilton New York Midtown, New York, New York, United States AAAI-2019 Hilton Hawaiian Village, Honolulu, Hawaii, United States AAAI-2018 Hilton New Orleans Riverside, New Orleans, Louisiana, United States AAAI-2017 San Francisco, California, United States AAAI-2016 Phoenix, Arizona, United States AAAI-2015 Austin, Texas, United States AAAI-2014 Québec Convention Center, Québec City, Québec, Canada AAAI-2013 Bellevue, Washington, United States AAAI-2012 Toronto, Ontario, Canada AAAI-2011 San Francisco, California, United States AAAI-2010 Westin Peachtree Plaza, Atlanta, Georgia, United States AAAI-2008 Chicago, Illinois, United States AAAI-2007 Toronto, Ontario, Canada AAAI-2006 Boston, Massachusetts, United States AAAI-2005 Pittsburgh, Pennsylvania, United States AAAI-2004 San Jose, California, United States AAAI-2002 Shaw conference center in Edmonton, Alberta, Canada AAAI-2000 Austin, Texas, United States AAAI-1999 Orlando, Florida, United States AAAI-1998 Madison, Wisconsin, United States AAAI-1997 Providence, Rhode Island, United States AAAI-1996 Portland, Oregon, United States AAAI-1994 Seattle, Washington, United States AAAI-1993 Washington Convention Center, Washington, D.C., United States AAAI-1992 San Jose Convention Center, San Jose, California, United States AAAI-1991 Anaheim Convention Center, Anaheim, California, United States AAAI-1990 Boston, Massachusetts, United States AAAI-1988 Saint Paul, Minnesota, United States AAAI-1987 Seattle, Washington, United States AAAI-1986 Philadelphia, Pennsylvania, United States AAAI-1984 University of Texas, Austin, Texas, United States AAAI-1983 Washington, D.C., United States AAAI-1982 Carnegie Mellon University and the University of Pittsburgh, Pittsburgh, Pennsylvania, United States AAAI-1980 Stanford, California, United States
Fuzzy pay-off method for real option valuation
The fuzzy pay-off method for real option valuation (FPOM or pay-off method) is a method for valuing real options, developed by Mikael Collan, Robert Fullér, and József Mezei; and published in 2009. It is based on the use of fuzzy logic and fuzzy numbers for the creation of the possible pay-off distribution of a project (real option). The structure of the method is similar to the probability theory based Datar–Mathews method for real option valuation, but the method is not based on probability theory and uses fuzzy numbers and possibility theory in framing the real option valuation problem. == Method == The Fuzzy pay-off method derives the real option value from a pay-off distribution that is created by using three or four cash-flow scenarios (most often created by an expert or a group of experts). The pay-off distribution is created simply by assigning each of the three cash-flow scenarios a corresponding definition with regards to a fuzzy number (triangular fuzzy number for three scenarios and a trapezoidal fuzzy number for four scenarios). This means that the pay-off distribution is created without any simulation whatsoever. This makes the procedure easy and transparent. The scenarios used are a minimum possible scenario (the lowest possible outcome), the maximum possible scenario (the highest possible outcome) and a best estimate (most likely to happen scenario) that is mapped as a fully possible scenario with a full degree of membership in the set of possible outcomes, or in the case of four scenarios used - two best estimate scenarios that are the upper and lower limit of the interval that is assigned a full degree of membership in the set of possible outcomes. The main observations that lie behind the model for deriving the real option value are the following: The fuzzy NPV of a project is (equal to) the pay-off distribution of a project value that is calculated with fuzzy numbers. The mean value of the positive values of the fuzzy NPV is the "possibilistic" mean value of the positive fuzzy NPV values. Real option value, ROV, calculated from the fuzzy NPV is the "possibilistic" mean value of the positive fuzzy NPV values multiplied with the positive area of the fuzzy NPV over the total area of the fuzzy NPV. The real option formula can then be written simply as: R O V = A ( P o s ) A ( P o s ) + A ( N e g ) × E [ A + ] {\displaystyle \mathrm {ROV} ={\frac {A(\mathrm {Pos} )}{A(\mathrm {Pos} )+A(\mathrm {Neg} )}}\times E[A_{+}]} where A(Pos) is the area of the positive part of the fuzzy distribution, A(Neg) is the area of the negative part of the fuzzy distribution, and E[A+] is the mean value of the positive part of the distribution. It can be seen that when the distribution is totally positive, the real options value reduces to the expected (mean) value, E[A+]. As can be seen, the real option value can be derived directly from the fuzzy NPV, without simulation. At the same time, simulation is not an absolutely necessary step in the Datar–Mathews method, so the two methods are not very different in that respect. But what is totally different is that the Datar–Mathews method is based on probability theory and as such has a very different foundation from the pay-off method that is based on possibility theory: the way that the two models treat uncertainty is fundamentally different. == Use of the method == The pay-off method for real option valuation is very easy to use compared to the other real option valuation methods and it can be used with the most commonly used spreadsheet software without any add-ins. The method is useful in analyses for decision making regarding investments that have an uncertain future, and especially so if the underlying data is in the form of cash-flow scenarios. The method is less useful if optimal timing is the objective. The method is flexible and accommodates easily both one-stage investments and multi-stage investments (compound real options). The method has been taken into use in some large international industrial companies for the valuation of research and development projects and portfolios. In these analyses triangular fuzzy numbers are used. Other uses of the method so far are, for example, R&D project valuation IPR valuation, valuation of M&A targets and expected synergies, valuation and optimization of M&A strategies, valuation of area development (construction) projects, valuation of large industrial real investments. The use of the pay-off method is lately taught within the larger framework of real options, for example at the Lappeenranta University of Technology and at the Tampere University of Technology in Finland.
CADE ATP System Competition
The CADE ATP System Competition (CASC) is an annual competition of fully automated theorem provers for classical logic. == Competition == CASC is associated with the Conference on Automated Deduction and the International Joint Conference on Automated Reasoning organized by the Association for Automated Reasoning. It has inspired similar competition in related fields, in particular the successful SMT-COMP competition for satisfiability modulo theories, the SAT Competition for propositional reasoners, and the modal logic reasoning competition. The first CASC, CASC-13, was held as part of the 13th Conference on Automated Deduction at Rutgers University, New Brunswick, NJ, in 1996. Among the systems competing were Otter and SETHEO.
D/Vision Pro
D/Vision Pro was one of the earliest marketed non-linear editing systems. It was released by TouchVision Systems, Inc. in the mid-1990s. The program was DOS-based and worked on either Intel's 386 or 486 processor. The system used AVI compression and worked with the Action Media II board. The system allowed users to digitize video, audio, and timecode, create an edit decision list (EDL), instantly play back the edited program, and output the finished EDL in a wide variety of formats. These cost-effective editing systems were used by numerous independent filmmakers and in low-budget productions during the mid-late 1990s. D/Vision Pro's low-quality compression led TouchVision (later renamed D/Vision Systems) to abandon it in favor of D/Vision Online, which was purchased by Discreet Logic and renamed edit. In June 2002, Discreet discontinued edit, as they did not want it to interfere with smoke sales which were more profitable. Discreet was later purchased by Autodesk.
Semantic Scholar
Semantic Scholar is a research tool for scientific literature. It is developed at the Allen Institute for AI and was publicly released in November 2015. Semantic Scholar uses modern techniques in natural language processing to support the research process, for example by providing automatically generated summaries of scholarly papers. The Semantic Scholar team is actively researching the use of artificial intelligence in natural language processing, machine learning, human–computer interaction, and information retrieval. Semantic Scholar began as a database for the topics of computer science, geoscience, and neuroscience. In 2017, the system began including biomedical literature in its corpus. As of September 2022, it includes over 200 million publications from all fields of science. == Technology == Semantic Scholar provides a one-sentence summary of scientific literature. One of its aims was to address the challenge of reading numerous titles and lengthy abstracts on mobile devices. It also seeks to ensure that the three million scientific papers published yearly reach readers, since it is estimated that only half of this literature is ever read. Artificial intelligence is used to capture the essence of a paper, generating it through an "abstractive" technique. The project uses a combination of machine learning, natural language processing, and machine vision to add a layer of semantic analysis to the traditional methods of citation analysis, and to extract relevant figures, tables, entities, and venues from papers. Another key AI-powered feature is Research Feeds, an adaptive research recommender that uses AI to quickly learn what papers users care about reading and recommends the latest research to help scholars stay up to date. It uses a paper embedding model trained using contrastive learning to find papers similar to those in each Library folder. Semantic Scholar also offers Semantic Reader, an augmented reader with the potential to revolutionize scientific reading by making it more accessible and richly contextual. Semantic Reader provides in-line citation cards that allow users to see citations with TLDR (short for Too Long, Didn't Read) automatically generated short summaries as they read and skimming highlights that capture key points of a paper so users can digest faster. In contrast with Google Scholar and PubMed, Semantic Scholar is designed to highlight the most important and influential elements of a paper. The AI technology is designed to identify hidden connections and links between research topics. Like the previously cited search engines, Semantic Scholar also exploits graph structures, which include the Microsoft Academic Knowledge Graph, Springer Nature's SciGraph, and the Semantic Scholar Corpus (originally a 45 million papers corpus in computer science, neuroscience and biomedicine). == Article identifier == Each paper hosted by Semantic Scholar is assigned a unique identifier called the Semantic Scholar Corpus ID (abbreviated S2CID). The following entry is an example: Liu, Ying; Gayle, Albert A; Wilder-Smith, Annelies; Rocklöv, Joacim (March 2020). "The reproductive number of COVID-19 is higher compared to SARS coronavirus". Journal of Travel Medicine. 27 (2). doi:10.1093/jtm/taaa021. PMID 32052846. S2CID 211099356. == Indexing == Semantic Scholar is free to use and unlike similar search engines (e.g., Google Scholar) does not search for material that is behind a paywall. One study compared the index scope of Semantic Scholar to Google Scholar, and found that for the papers cited by secondary studies in computer science, the two indices had comparable coverage, each only missing a handful of the papers. == Number of users and publications == As of January 2018, following a 2017 project that added biomedical papers and topic summaries, the Semantic Scholar corpus included more than 40 million papers from computer science and biomedicine. In March 2018, Doug Raymond, who developed machine learning initiatives for the Amazon Alexa platform, was hired to lead the Semantic Scholar project. As of August 2019, the number of included papers metadata (not the actual PDFs) had grown to more than 173 million after the addition of the Microsoft Academic Graph records. In 2020, a partnership between Semantic Scholar and the University of Chicago Press Journals made all articles published under the University of Chicago Press available in the Semantic Scholar corpus. At the end of 2020, Semantic Scholar had indexed 190 million papers. In 2020, Semantic Scholar reached seven million users per month.