检索结果-内蒙古大学图书馆

Beyond Black-Box Advice: Learning-Augmented Algorithms for MDPs with Q-Value Predictions 37

学校读者我要写书评

暂无评论

Beyond Black-Box Advice: Learning-Augmented Algorithms for M...

37th Conference on Neural Information Processing Systems, NeurIPS 2023

作者： Li, Tongxin Lin, Yiheng Ren, Shaolei Wierman, Adam School of Data Science CUHK-SZ China Computing + Mathematical Sciences Caltech United States Electrical & Computer Engineering UC Riverside United States

ISBN: (纸本)9781713899921

We study the tradeoff between consistency and robustness in the context of a single-trajectory time-varying Markov Decision Process (MDP) with untrusted machine-learned advice. Our work departs from the typical approach of treating advice as coming from black-box sources by instead considering a setting where additional information about how the advice is generated is available. We prove a first-of-its-kind consistency and robustness tradeoff given Q-value advice under a general MDP model that includes both continuous and discrete state/action spaces. Our results highlight that utilizing Q-value advice enables dynamic pursuit of the better of machine-learned advice and a robust baseline, thus result in near-optimal performance guarantees, which provably improves what can be obtained solely with black-box advice. © 2023 Neural information processing systems foundation. All rights reserved.

关键词： Markov processes

What Did My Car Say? Impact of Autonomous Vehicle Explanation Errors and Driving Context on Comfort, Reliance, Satisfaction, and Driving Confidence 25

学校读者我要写书评

暂无评论

What Did My Car Say? Impact of Autonomous Vehicle Explanatio...

2025 CHI Conference on Human Factors in Computing Systems, CHI 2025

作者： Kaufman, Robert A Broukhim, Aaron Kirsh, David Weibel, Nadir Cognitive Science University of California San Diego CA United States Computer Science & Engineering University of California San Diego San Diego CA United States Computer Science and Engineering The Design Lab UC San Diego San Diego CA United States

ISBN: (纸本)9798400713941

Explanations for autonomous vehicle (AV) decisions may build trust, however, explanations can contain errors. In a simulated driving study (n = 232), we tested how AV explanation errors, driving context characteristics (perceived harm and driving difficulty), and personal traits (prior trust and expertise) affected a passenger's comfort in relying on an AV, preference for control, confidence in the AV's ability, and explanation satisfaction. Errors negatively affected all outcomes. Surprisingly, despite identical driving, explanation errors reduced ratings of the AV's driving ability. Severity and potential harm amplified the negative impact of errors. Contextual harm and driving difficulty directly impacted outcome ratings and influenced the relationship between errors and outcomes. Prior trust and expertise were positively associated with outcome ratings. Results emphasize the need for accurate, contextually adaptive, and personalized AV explanations to foster trust, reliance, satisfaction, and confidence. We conclude with design, research, and deployment recommendations for trustworthy AV explanation systems. © 2025 Copyright held by the owner/author(s).

关键词： AI Errors Autonomous Vehicles Explainable AI

Slug Mobile: Test-Bench for RL Testing

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Morris, Jonathan Wellington Shah, Vishrut Besanceney, Alex Shah, Daksh Gilpin, Leilani H. UC Santa Cruz United States Department of Computer Science and Engineering UC Santa Cruz United States

Sim-to real gap in Reinforcement Learning is when a model trained in a simulator does not translate to the real world. This is a problem for Autonomous Vehicles (AVs) as vehicle dynamics can vary from simulation to reality, and also from vehicle to vehicle. Slug Mobile is a one tenth scale autonomous vehicle created to help address the sim-to-real gap for AVs by acting as a test-bench to develop models that can easily scale from one vehicle to another. In addition to traditional sensors found in other one tenth scale AVs, we have also included a Dynamic Vision Sensor so we can train Spiking Neural Networks running on neuromorphic hardware. © 2024, CC BY-SA.

关键词： Reinforcement learning

White-box logic obfuscation: A Transparent Solution to Hardware Piracy and Reverse engineering 25

学校读者我要写书评

暂无评论

White-box logic obfuscation: A Transparent Solution to Hardw...

Proceedings of the 30th Asia and South Pacific Design Automation Conference

作者： Leon Li Alex Orailoglu Department of Computer Science and Engineering UC San Diego La Jolla California USA

ISBN: (纸本)9798400706356

Reverse engineering the functional specification from a netlist is a challenging task that enables IP piracy and tampering. Traditional logic locking techniques, which depend on external activation with secrets stored in tamper-proof locations to thwart reverse engineering, have repeatedly been compromised by key recovery attacks. Their vulnerability highlights the flawed assumption of relying on tamper-proof secrets for building hardware security solutions, especially when activated devices are deployed in open environments where they are exposed to attackers for functional queries and probing. This paper presents white-box logic obfuscation (WBLO) as a novel solution to safeguard control logic from functional reverse engineering, even when attackers have full visibility of the operational netlist. WBLO eliminates the need for post-manufacturing activation and reliance on tamper-proof key storage by securing the design with keys that are autonomously updated internally through the legitimate sequential execution of the device. The proposed approach invalidates functional analysis under arbitrary probing and combinational queries. We examine the implementation challenges inherent in the WBLO process and identify critical design considerations that enhance security and efficiency. Building on these insights, we suggest a prioritization for various design transformation and synthesis rules that achieve robust security in the white-box attack model while minimizing implementation overheads.

关键词： hardware security

Stability-Aware Training of Machine Learning Force Fields with Differentiable Boltzmann Estimators

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Raja, Sanjeev Amin, Ishan Pedregosa, Fabian Krishnapriyan, Aditi S. Department of Computer Science UC Berkeley United States Department of Computer Science Department of Physics UC Berkeley United States Google Deepmind United Kingdom Department of Computer Science Department of Chemical Engineering UC Berkeley LBNL United States

Machine learning force fields (MLFFs) are an attractive alternative to ab-initio methods for molecular dynamics (MD) simulations. However, they can produce unstable simulations, limiting their ability to model phenomena occurring over longer timescales and compromising the quality of estimated observables. To address these challenges, we present Stability-Aware Boltzmann Estimator (StABlE) Training, a multi-modal training procedure which leverages joint supervision from reference quantum-mechanical calculations and system observables. StABlE Training iteratively runs many MD simulations in parallel to seek out unstable regions, and corrects the instabilities via supervision with a reference observable. We achieve efficient end-to-end automatic differentiation through MD simulations using our Boltzmann Estimator, a generalization of implicit differentiation techniques to a broader class of stochastic algorithms. Unlike existing techniques based on active learning, our approach requires no additional ab-initio energy and forces calculations to correct instabilities. We demonstrate our methodology across organic molecules, tetrapeptides, and condensed phase systems, using three modern MLFF architectures. StABlE-trained models achieve significant improvements in simulation stability, data efficiency, and agreement with reference observables. Crucially, the stability improvements cannot be matched by simply reducing the simulation timestep, meaning that StABlE Training effectively allows for larger timesteps in MD simulations. By incorporating observables into the training process alongside first-principles calculations, StABlE Training can be viewed as a general semi-empirical framework applicable across MLFF architectures and systems. This makes it a powerful tool for training stable and accurate MLFFs, particularly in the absence of large reference datasets. Our code is publicly available at https://***/ASK-Berkeley/StABlE-Training. © 2024, CC BY.

关键词： Stochastic systems

Strong bounds for skew corner-free sets

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Jaber, Michael Lovett, Shachar Ostuni, Anthony Department of Computer Science UT Austin Australia Department of Computer Science and Engineering UC San Diego United States

Motivated by applications to matrix multiplication algorithms, Pratt asked (ITCS’24) how large a subset of [n] × [n] could be without containing a skew-corner: three points (x, y), (x, y + h), (x + h, y1) with h ≠ 0. We prove any skew corner-free set has size at most exp(-Ωplog1/12 n)) ∙ n2, nearly matching the best known lower bound of exp(-O(√log n)) ∙ n2 by Beker (arXiv’24). Our techniques generalize those of Kelley and Meka’s recent breakthrough on three-term arithmetic progression (FOCS’23), answering a question of Beker (arXiv’24). © 2024, CC BY.

关键词：

Reducing the Carbon Footprint of EdTech with Repurposed Devices

学校读者我要写书评

暂无评论

Reducing the Carbon Footprint of EdTech with Repurposed Devi...

International Green Computing Conference

作者： Jennifer Switzer Subash Katel Jaemok Christian Lee Ashwin Rohit Alagiri Rajan Ryan Kastner Pat Pannuto Computer Science and Engineering UC San Diego La Jolla CA USA

ISBN: (数字)9798331507862

ISBN: (纸本)9798331507879

Education technology (EdTech) is an important tool for streamlining and improving course administration and teaching. Many modern EdTech tools rely on cloud services to host containerized applications. While this is convenient, it is also costly in terms of both dollars and carbon *** propose the alternative approach of hosting containerized EdTech applications on local clusters of upcycled Android devices. We perform an evaluation of the Google Pixel Fold for handling educational workloads. Our findings suggest that such repurposed device could effectively bridge the gap between mobile and traditional computing platforms in education, open new avenues for accessible educational computing environments.

关键词： Performance evaluation Education Green products Carbon dioxide Internet Carbon footprint

A NEAR-OPTIMAL ALGORITHM FOR SHORTEST PATHS AMONG CURVED OBSTACLES IN THE PLANE

学校读者我要写书评

暂无评论

SIAM Journal on Computing 2022年第4期51卷 1296-1340页

作者： Hershberger, John Suri, Subhash Yildiz, Hakan Siemens EDA WilsonvilleOR97070 United States Department of Computer Science UC Santa Barbara Santa BarbaraCA93106 United States Department of Computer Engineering Middle East Technical University Ankara Turkey

We propose an algorithm for the problem of computing shortest paths among curved obstacles in the plane. If the obstacles have O(n) description complexity, then the algorithm runs in O(n log n) time plus a term dependent on the properties of the boundary arcs. Specifically, if the arcs allow a certain kind of bisector intersection to be computed in constant time, or even in O(log n) time, then the running time of the overall algorithm is O(n log n). If the arcs support only constant-time tangent, intersection, and length queries, as is customarily assumed, then the algorithm computes an approximate shortest path, with relative error Ε, in time O(n log n + n log 1\e ). In fact, the algorithm computes an approximate shortest path map, a data structure with O(n log n) size, that allows it to report the (approximate) length of a shortest path from a fixed source point to any query point in the plane in O(log n) time. By applying an idea due to Wang [Proceedings of the 32nd Annual ACM-SIAM Symposium on Discrete Algorithms, 2021, pp. 810-821], the algorithm's working storage and the size of the approximate shortest path map can be reduced to O(n). © 2022 Society for Industrial and Applied Mathematics.

关键词： Digital storage

Fast, optimal, and dynamic electoral campaign budgeting by a generalized Colonel Blotto game

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Valles, Thomas Beaglehole, Daniel Department of Mathematics UC San Diego United States Computer Science and Engineering UC San Diego United States

The Colonel Blotto game is a deeply studied theoretical model for competitive allocation environments including elections, advertising, and ecology. However, the original formulation of Colonel Blotto has had few practical implications due to the lack of fast algorithms to compute its optimal strategies and the limited applicability of its winner-take-all reward distribution. We demonstrate that the Colonel Blotto game can be a practical model for competitive allocation environments by implementing the multiplicative weights update algorithm from Beaglehole et al. (2023). In particular, using that this algorithm allows for arbitrary winning-rules, we study strategies for a more realistic model of political campaigning we term Electoral Colonel Blotto. Contrary to existing theory and the implemented allocation strategies from U.S. presidential elections, we find that the optimal response to Democratic and Republican strategies in the 2008 and 2020 presidential elections was to focus allocations on a subset of states and sacrifice winning probability on others. We also found that campaigners should compete for undecided voters even in states where the opponent has significantly many more decided voters. © 2024, CC0.

关键词： Budget control