检索结果-内蒙古大学图书馆

Randomized Entity-wise Factorization for Multi-Agent Reinforcement Learning 38

学校读者我要写书评

暂无评论

Randomized Entity-wise Factorization for Multi-Agent Reinfor...

38th International Conference on Machine Learning, ICML 2021

作者： Iqbal, Shariq Schroeder de Witt, Christian A. Peng, Bei Böhmer, Wendelin Whiteson, Shimon Sha, Fei Department of Computer Science University of Southern California United States Department of Computer Science University of Oxford United Kingdom Department of Software Technology Delft University of Technology Netherlands Google Research

ISBN: (纸本)9781713845065

Multi-agent settings in the real world often involve tasks with varying types and quantities of agents and non-agent entities;however, common patterns of behavior often emerge among these agents/entities. Our method aims to leverage these commonalities by asking the question: "What is the expected utility of each agent when only considering a randomly selected subgroup of its observed entities?" By posing this counterfactual question, we can recognize state-action trajectories within sub-groups of entities that we may have encountered in another task and use what we learned in that task to inform our prediction in the current one. We then reconstruct a prediction of the full returns as a combination of factors considering these disjoint groups of entities and train this "randomly factorized" value function as an auxiliary objective for value-based multi-agent reinforcement learning. By doing so, our model can recognize and leverage similarities across tasks to improve learning efficiency in a multi-task setting. Our approach, Randomized Entity-wise Factorization for Imagined Learning (R E F I L), outperforms all strong baselines by a significant margin in challenging multi-task StarCraft micromanagement settings. Copyright © 2021 by the author(s)

关键词： Factorization

Scaling and logic in the color code on a superconducting quantum processor

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Lacroix, Nathan Bourassa, Alexandre Heras, Francisco J.H. Zhang, Lei M. Bausch, Johannes Senior, Andrew W. Edlich, Thomas Shutty, Noah Sivak, Volodymyr Bengtsson, Andreas McEwen, Matt Higgott, Oscar Kafri, Dvir Claes, Jahan Morvan, Alexis Chen, Zijun Zalcman, Adam Madhuk, Sid Acharya, Rajeev Beni, Laleh Aghababaie Aigeldinger, Georg Alcaraz, Ross Andersen, Trond I. Ansmann, Markus Arute, Frank Arya, Kunal Asfaw, Abraham Atalaya, Juan Babbush, Ryan Ballard, Brian Bardin, Joseph C. Bilmes, Alexander Blackwell, Sam Bovaird, Jenna Bowers, Dylan Brill, Leon Broughton, Michael Browne, David A. Buchea, Brett Buckley, Bob B. Burger, Tim Burkett, Brian Bushnell, Nicholas Cabrera, Anthony Campero, Juan Chang, Hung-Shen Chiaro, Ben Chih, Liang-Ying Cleland, Agnetta Y. Cogan, Josh Collins, Roberto Conner, Paul Courtney, William Crook, Alexander L. Curtin, Ben Das, Sayan Demura, Sean De Lorenzo, Laura Di Paolo, Agustin Donohoe, Paul Drozdov, Ilya Dunsworth, Andrew Eickbusch, Alec Elbag, Aviv Moshe Elzouka, Mahmoud Erickson, Catherine Ferreira, Vinicius S. Burgos, Leslie Flores Forati, Ebrahim Fowler, Austin G. Foxen, Brooks Ganjam, Suhas Garcia, Gonzalo Gasca, Robert Genois, Élie Giang, William Gilboa, Dar Gosula, Raja Dau, Alejandro Grajales Graumann, Dietrich Greene, Alex Gross, Jonathan A. Ha, Tan Habegger, Steve Hansen, Monica Harrigan, Matthew P. Harrington, Sean D. Heslin, Stephen Heu, Paula Hiltermann, Reno Hilton, Jeremy Hong, Sabrina Huang, Hsin-Yuan Huff, Ashley Huggins, William J. Jeffrey, Evan Jiang, Zhang Jin, Xiaoxuan Joshi, Chaitali Juhas, Pavol Kabel, Andreas Kang, Hui Karamlou, Amir H. Kechedzhi, Kostyantyn Khaire, Trupti Khattar, Tanuj Khezri, Mostafa Kim, Seon Klimov, Paul V. Kobrin, Bryce Korotkov, Alexander N. Kostritsa, Fedor Kreikebaum, John Mark Kurilovich, Vladislav D. Landhuis, David Lange-Dei, Tiano Langley, Brandon W. Laptev, Pavel Lau, Kim-Ming Ledford, Justin Lee, Kenny Lester, Brian J. Le Guevel, Loïck Li, Wing Yan Li, Yin Lill, Alexander T. Livingston, William P. Locharla, Aditya Lucero, Erik L Google Research Mountain ViewCA United States Department of Physics ETH Zurich Switzerland Google DeepMind London United Kingdom Department of Electrical and Computer Engineering University of Massachusetts AmherstMA United States Department of Physics University of Connecticut StorrsCT United States Department of Computer Science University of California Santa BarbaraCA United States Research Laboratory of Electronics Massachusetts Institute of Technology CambridgeMA United States Department of Electrical Engineering and Computer Science Massachusetts Institute of Technology CambridgeMA United States Department of Physics Massachusetts Institute of Technology CambridgeMA United States

Quantum error correction [1-6] is essential for bridging the gap between the error rates of physical devices and the extremely low logical error rates required for quantum algorithms. Recent error-correction demonstrations on superconducting processors [7-13] have focused primarily on the surface code [14], which offers a high error threshold but poses limitations for logical operations. In contrast, the color code [15] enables much more efficient logic, although it requires more complex stabilizer measurements and decoding techniques. Measuring these stabilizers in planar architectures such as superconducting qubits is challenging, and so far, realizations of color codes [16-24] have not addressed performance scaling with code size on any platform. Here, we present a comprehensive demonstration of the color code on a superconducting processor [12], achieving logical error suppression and performing logical operations. Scaling the code distance from three to five suppresses logical errors by a factor of Λ3/5 = 1.56(4). Simulations indicate this performance is below the threshold of the color code, and furthermore that the color code may be more efficient than the surface code with modest device improvements. Using logical randomized benchmarking [25], we find that transversal Clifford gates add an error of only 0.0027(3), which is substantially less than the error of an idling error correction cycle. We inject magic states [26], a key resource for universal computation, achieving fidelities exceeding 99 % with post-selection (retaining about 75 % of the data). Finally, we successfully teleport logical states between distance-three color codes using lattice surgery [27], with teleported state fidelities between 86.5(1) % and 90.7(1) %. This work establishes the color code as a compelling research direction to realize fault-tolerant quantum computation on superconducting processors in the near future. © 2024, CC BY.

关键词： Color codes

Midgame Attacks and Defense Against Them 7th

学校读者我要写书评

暂无评论

Midgame Attacks and Defense Against Them

7th International Symposium on Cyber Security, Cryptology, and Machine Learning, CSCML 2023

作者： Chang, Donghoon Yung, Moti National Institute of Standards and Technology GaithersburgMD United States Strativia LargoMD United States Delhi India Google Inc. Mountain View United States Department of Computer Science Columbia University New York United States

ISBN: (纸本)9783031346705

In this paper, we propose the Midgame Security attack model, where it is assumed that at some point in the middle of computation with a secret key, and after some secure work (typically but not necessarily initial one), the powerful adversary sees the entire internal state and attempts key recovery/forgery. This security model is motivated by a few trends: First and primarily, it may represent a model in which part of the computation is done in a possibly insecure environment (e.g., the emerging modes of cloud server delegation, hosting environment, general pc, the cloud, etc.), where the insecure environment performs the bulk of the work, after some initial or intermediate (relatively small amount of) work at a trusted location which holds the cryptographic keys (client, co-processor, trusted hardware with leakage countermeasures, an enclave in the cloud, etc.). Secondly, from a leakage perspective, the model represents a total leakage in the system at some point after some secured work has been done without leakage (perhaps at a different location). The model is novel (though, superficially, it has a flavor of forward security), and is most meaningful to demonstrate exposures of constructions where there is an obvious lengthy progress of computation (e.g., MACing or (Authenticated) Encrypting of long messages) which is done without the cryptographic keys present, and when we want short usage of keys (to minimize their exposure). In these cases, initially in secure periods the key may be blended into the state of the computation and an attacker task is to recover that key in spite of the blending from the leakage from the environment which never hold any key. We employ the new model to analyze numerous concrete cryptosystems and mainly find key recovery or forgery attacks. We first compare HMAC based on the SHA-3 finalists in this new midgame security model. One thing we show is that the domain extension of Keccak, called the sponge construction, is exposed in a HM

关键词： Blending

Auto-Linear Phenomenon in Subsurface Imaging 41

学校读者我要写书评

暂无评论

Auto-Linear Phenomenon in Subsurface Imaging

41st International Conference on Machine Learning, ICML 2024

作者： Feng, Yinan Chen, Yinpeng Jin, Peng Feng, Shihang Lin, Youzuo Department of Computer Science The University of North Carolina Chapel Hill United States Google Research United States College of Information Sciences and Technology The Pennsylvania State University United States Earth and Environmental Sciences Division Los Alamos National Laboratory United States School of Data Science and Society The University of North Carolina Chapel Hill United States

Subsurface imaging involves solving full waveform inversion (FWI) to predict geophysical properties from measurements. This problem can be reframed as an image-to-image translation, with the usual approach being to train an encoder-decoder network using paired data from two domains: geophysical property and measurement. A recent seminal work (InvLINT) demonstrates there is only a linear mapping between the latent spaces of the two domains, and the decoder requires paired data for training. This paper extends this direction by demonstrating that only linear mapping necessitates paired data, while both the encoder and decoder can be learned from their respective domains through self-supervised learning. This unveils an intriguing phenomenon (named Auto-Linear) where the self-learned features of two separate domains are automatically linearly correlated. Compared with existing methods, our Auto-Linear has four advantages: (a) solving both forward and inverse modeling simultaneously, (b) applicable to different subsurface imaging tasks and achieving markedly better results than previous methods, (c)enhanced performance, especially in scenarios with limited paired data and in the presence of noisy data, and (d) strong generalization ability of the trained encoder and decoder. Copyright 2024 by the author(s)

关键词： Subsalt imaging

How Many Ratings per Item are Necessary for Reliable Significance Testing?

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Homan, Christopher M. Korn, Flip Welty, Chris Department of Computer Science Rochester Institute of Technology RochesterNY14607 United States Google Research New YorkNY10011 United States

Most approaches to machine learning evaluation assume that machine and human responses are repeatable enough to be measured against data with unitary, authoritative, "gold standard" responses, via simple metrics such as accuracy, precision, and recall that assume scores are independent given the test item. However, AI models have multiple sources of stochasticity and the human raters who create gold standards tend to disagree with each other, often in meaningful ways, hence a single output response per input item may not provide enough information. We introduce methods for determining whether an (existing or planned) evaluation dataset has enough responses per item to reliably compare the performance of one model to another. We apply our methods to several of very few extant gold standard test sets with multiple disaggregated responses per item and show that there are usually not enough responses per item to reliably compare the performance of one model against another. Our methods also allow us to estimate the number of responses per item for hypothetical datasets with similar response distributions to the existing datasets we study. When two models are very far apart in their predictive performance, fewer raters are needed to confidently compare them, as expected. However, as the models draw closer, we find that a larger number of raters than are currently typical in annotation collection are needed to ensure that the power analysis correctly reflects the difference in performance. © 2024, CC BY.

关键词： Stochastic models

Simplicity Bias in 1-Hidden Layer Neural Networks

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Morwani, Depen Batra, Jatin Jain, Prateek Netrapalli, Praneeth Department of Computer Science Harvard University CambridgeMA United States School of Technology and Computer Science TIFR Mumbai India Google Research Bengaluru India Alphabetical ordering

Recent works (Shah et al., 2020;Chen et al., 2021) have demonstrated that neural networks exhibit extreme simplicity bias (SB). That is, they learn only the simplest features to solve a task at hand, even in the presence of other, more robust but more complex features. Due to the lack of a general and rigorous definition of features, these works showcase SB on semi-synthetic datasets such as Color-MNIST, MNIST-CIFAR where defining features is relatively easier. In this work, we rigorously define as well as thoroughly establish SB for one hidden layer neural networks. More concretely, (i) we define SB as the network essentially being a function of a low dimensional projection of the inputs (ii) theoretically, we show that when the data is linearly separable, the network primarily depends on only the linearly separable (1-dimensional) subspace even in the presence of an arbitrarily large number of other, more complex features which could have led to a significantly more robust classifier, (iii) empirically, we show that models trained on real datasets such as Imagenette and Waterbirds-Landbirds indeed depend on a low dimensional projection of the inputs, thereby demonstrating SB on these datasets, iv) finally, we present a natural ensemble approach that encourages diversity in models by training successive models on features not used by earlier models, and demonstrate that it yields models that are significantly more robust to Gaussian noise. Copyright © 2023, The Authors. All rights reserved.

关键词： Complex networks

Insufficient Statistics Perturbation: Stable Estimators for Private Least Squares

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Brown, Gavin Hayase, Jonathan Hopkins, Samuel Kong, Weihao Liu, Xiyang Oh, Sewoong Perdomo, Juan C. Smith, Adam Paul G. Allen School of Computer Science and Engineering University of Washington United States Department of Electrical Engineering and Computer Science Massachusetts Institute of Technology United States Google Research United States Harvard University United States Department of Computer Science Boston University United States

We present a sample- and time-efficient differentially private algorithm for ordinary least squares, with error that depends linearly on the dimension and is independent of the condition number of XX, where X is the design matrix. All prior private algorithms for this task require either d3/2 examples, error growing polynomially with the condition number, or exponential time. Our near-optimal accuracy guarantee holds for any dataset with bounded statistical leverage and bounded residuals. Technically, we build on the approach of Brown et al. (2023) for private mean estimation, adding scaled noise to a carefully designed stable nonprivate estimator of the empirical regression vector. Copyright © 2024, The Authors. All rights reserved.

关键词： Machine learning

Mirror Descent Algorithms with Nearly Dimension-Independent Rates for Differentially-Private Stochastic Saddle-Point Problems

学校读者我要写书评

暂无评论

arXiv 2024年

作者： González, Tomás Guzmán, Cristóbal Paquette, Courtney Machine Learning Department School of Computer Science Carnegie Mellon University United States Institute for Mathematical and Computational Engineering Faculty of Mathematics School of Engineering Pontificia Universidad Católica de Chile Chile Department of Mathematics and Statistics McGill University Google Deepmind Canada

We study the problem of differentially-private (DP) stochastic (convex-concave) saddle-points in the polyhedral setting. We propose (Ε, δ)-DP algorithms based on stochastic mirror descent that attain nearly dimension-independent convergence rates for the expected duality gap, a type of guarantee that was known before only for bilinear objectives. For convex-concave and first-order-smooth stochastic objectives, our algorithms attain a rate of (Formula Presented), where d is the dimension of the problem and n the dataset size. Under an additional second-order-smoothness assumption, we improve the rate on the expected gap to (Formula Presented). Under this additional assumption, we also show, by using bias-reduced gradient estimators, that the duality gap is bounded by (Formula Presented) with constant success probability. This result provides evidence of the near-optimality of the approach. Finally, we show that combining our methods with acceleration techniques from online learning leads to the first algorithm for DP Stochastic Convex Optimization in the polyhedral setting that is not based on Frank-Wolfe methods. For convex and first-order-smooth stochastic objectives, our algorithms attain an excess risk of (Formula Presented), and when additionally assuming second-order-smoothness, we improve the rate to plog(d)/n + log(d)/√nΕ. Instrumental to all of these results are various extensions of the classical Maurey Sparsification Lemma, which may be of independent interest. © 2024, CC BY.

关键词： Mirrors

Decoding Visual Experience and Mapping Semantics through Whole-Brain Analysis Using fMRI Foundation Models

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Wang, Yanchen Turnbull, Adam Xiang, Tiange Xu, Yunlong Zhou, Sa Masoud, Adnan Azizi, Shekoofeh Lin, Feng Vankee Adeli, Ehsan Department of Psychiatry and Behavioral Sciences Stanford University StanfordCA United States Department of Computer Science Stanford University StanfordCA United States Department of Neurobiology University of Chicago ChicagoIL United States UST TampaFL United States Google DeepMind TorontoON Canada

Neural decoding, the process of understanding how brain activity corresponds to different stimuli, has been a primary objective in cognitive sciences. Over the past three decades, advancements in functional Magnetic Resonance Imaging (fMRI) and machine learning have greatly improved our ability to map visual stimuli to brain activity, especially in the visual cortex. Concurrently, research has expanded into decoding more complex processes like language and memory across the whole brain, utilizing techniques to handle greater variability and improve signal accuracy. We argue that "seeing" involves more than just mapping visual stimuli onto the visual cortex;it engages the entire brain, as various emotions and cognitive states can emerge from observing different scenes. In this paper, we develop algorithms to enhance our understanding of visual processes by incorporating whole-brain activation maps while individuals are exposed to visual stimuli. We utilize large-scale fMRI encoders and Image generative models (encoders & decoders) pre-trained on large public datasets, which are then fine-tuned through Image-fMRI contrastive learning. Our models hence can decode visual experience across the entire cerebral cortex, surpassing the traditional confines of the visual cortex. Using a public dataset (BOLD5000), we first compare our method with state-of-the-art approaches to decoding visual processing and show improved predictive semantic accuracy by 43%. A network ablation analysis suggests that beyond the visual cortex, the default mode network contributes most to decoding stimuli, in line with the proposed role of this network in sense-making and semantic processing. Additionally, we implemented zero-shot imagination decoding on an extra validation dataset, achieving a p-value of 0.0206 for mapping the reconstructed images and ground-truth text stimuli, which substantiates the model’s capability to capture semantic meanings across various scenarios. These findings undersc

关键词： Brain mapping