BEGIN:VCALENDAR VERSION:2.0 PRODID:-//IFDS - ECPv6.0.1.1//NONSGML v1.0//EN CALSCALE:GREGORIAN METHOD:PUBLISH X-WR-CALNAME:IFDS X-ORIGINAL-URL:https://ifds.info X-WR-CALDESC:Events for IFDS REFRESH-INTERVAL;VALUE=DURATION:PT1H X-Robots-Tag:noindex X-PUBLISHED-TTL:PT1H BEGIN:VTIMEZONE TZID:America/Los_Angeles BEGIN:DAYLIGHT TZOFFSETFROM:-0800 TZOFFSETTO:-0700 TZNAME:PDT DTSTART:20240310T100000 END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:-0700 TZOFFSETTO:-0800 TZNAME:PST DTSTART:20241103T090000 END:STANDARD TZID:America/Chicago BEGIN:DAYLIGHT TZOFFSETFROM:-0600 TZOFFSETTO:-0500 TZNAME:CDT DTSTART:20240310T080000 END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:-0500 TZOFFSETTO:-0600 TZNAME:CST DTSTART:20241103T070000 END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTART;TZID=America/Los_Angeles:20240216T133000 DTEND;TZID=America/Los_Angeles:20240216T143000 DTSTAMP:20260425T112336 CREATED:20240318T212625Z LAST-MODIFIED:20240318T212625Z UID:2884-1708090200-1708093800@ifds.info SUMMARY:Offline Multi-task Transfer RL with Representational Penalization DESCRIPTION:Speaker Bio: Avinandan is a second year PhD student\, advised by Maryam Fazel and Lillian Ratliff. His interests are in sequential learning and game theory. \n\n\nAbstract: We study the problem of representational transfer in offline Reinforcement Learning (RL)\, where a learner has access to episodic data from a number of source tasks collected a priori\, and aims to learn a shared representation to be used in finding a good policy for a target task. Unlike in online RL where the agent interacts with the environment while learning a policy\, in the offline setting there cannot be such interactions in either the source tasks or the target task; thus multi-task offline RL can suffer from incomplete coverage.We propose an algorithm to compute pointwise uncertainty measures for the learnt representation\, and establish a data-dependent upper bound for the suboptimality of the learnt policy for the target task. Our algorithm leverages the collective exploration done by source tasks to mitigate poor coverage at some points by a few tasks\, thus overcoming the limitation of needing uniformly good coverage for a meaningful transfer by existing offline algorithms. We complement our theoretical results with empirical evaluation on a rich-observation MDP which requires many samples for complete coverage. Our findings illustrate the benefits of penalizing and quantifying the uncertainty in the learnt representation. URL:https://ifds.info/event/offline-multi-task-transfer-rl-with-representational-penalization/ LOCATION:CSE (Allen) 403 CATEGORIES:MLOpt@UWash END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Los_Angeles:20240223T133000 DTEND;TZID=America/Los_Angeles:20240223T143000 DTSTAMP:20260425T112336 CREATED:20240318T212752Z LAST-MODIFIED:20240318T212752Z UID:2886-1708695000-1708698600@ifds.info SUMMARY:GumbelSpec Sampling for Accelerating LLM Inference DESCRIPTION:Bio: Tianxiao Shen is a postdoctoral scholar at the University of Washington\, working with Yejin Choi and Zaid Harchaoui. Her research interests lie in natural language processing and machine learning\, in particular developing models and algorithms for efficient\, accurate\, diverse\, flexible and controllable text generation. She received her PhD from MIT\, advised by Regina Barzilay and Tommi Jaakkola. Before that\, she did her undergrad at Tsinghua University. \n\n\n\n\nAbstract: We propose GumbelSpec sampling\, a novel algorithm that leverages smaller language models to accelerate inference of large language models without changing their output distribution. Central to our approach is the application of the Gumbel-Softmax technique to convert the stochastic decoding process into a deterministic process by integrating independently sampled Gumbel noise. Employing the same set of Gumbel noise\, we perform beam search on the smaller model to generate multiple candidate short continuations\, and then utilize tree-based attention to efficiently verify them in parallel using the larger model. GumbelSpec sampling significantly improves upon previous rejection sampling based speculative decoding methods by increasing the token acceptance rate by 1.7x-2.2x and achieving an additional speedup of 1.2x-1.5x. This results in a total speedup of 1.5x-2.6x compared to traditional autoregressive decoding. URL:https://ifds.info/event/gumbelspec-sampling-for-accelerating-llm-inference/ LOCATION:CSE (Allen) 403 CATEGORIES:MLOpt@UWash END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20240308T133000 DTEND;TZID=America/Chicago:20240308T143000 DTSTAMP:20260425T112336 CREATED:20240318T213102Z LAST-MODIFIED:20240318T213102Z UID:2890-1709904600-1709908200@ifds.info SUMMARY:Low-Rank Structures in Optimal Transport DESCRIPTION:Bio: Meyer Scetbon is currently a Research Scientist at Microsoft Research. He completed his PhD at Institut Polytechnique de Paris\, advised by M. Cuturi. He did\, as a visiting student\, his MS theses UW and Technion\, on kernel-based viewpoints on deep neural networks advised by Z. Harchaoui\, and end-to-end signal and image denoising advised by M. Elad\, respectively. \n\n\n\n\n\nAbstract: Optimal transport (OT) plays an increasingly important role in machine learning (ML) to compare probability distributions. Yet\, it poses\, in its original form\, several challenges when used for applied problems: (i) computing OT between discrete distributions amounts to solving a large and expensive network flow problem which requires a supercubic complexity in the number of points; (ii) estimating OT using sampled measures is doomed by the curse of dimensionality. These issues can be mitigated using an entropic regularization\, solved with the Sinkhorn algorithm\, which improves on both statistical and computational aspects. While much faster\, entropic OT still requires a quadratic complexity with respect to the number of points and therefore remains prohibitive for large-scale problems. In this talk\, I will present new regularization approaches for the OT problem\, as well as its quadratic extension\, the Gromov-Wasserstein (GW) problem\, which impose low-rank structures on the admissible couplings. This results in the development of new algorithms that enjoy a linear complexity both in time and memory with respect to the number of points\, enabling their applications in the large-scale setting where millions of points need to be compared. Additionally\, I will show that these new regularization schemes have better statistical performances compared to the entropic approach\, that they naturally interpolate between the Maximum Mean Discrepancy (MMD) and OT\, and that they offer general clustering methods for arbitrary geometry.Website:  URL:https://ifds.info/event/low-rank-structures-in-optimal-transport/ LOCATION:CSE (Allen) 403 CATEGORIES:MLOpt@UWash END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Los_Angeles:20240315T133000 DTEND;TZID=America/Los_Angeles:20240315T143000 DTSTAMP:20260425T112336 CREATED:20240318T212943Z LAST-MODIFIED:20240318T212943Z UID:2888-1710509400-1710513000@ifds.info SUMMARY:Optimized Decision Making via Active Learning of Stochastic Hamiltonians DESCRIPTION:Speaker: Prof. Chandrajit Bajaj \, UT Austin \nAbstract: A Hamiltonian represents the energy of a dynamical system in phase space with coordinates of position and momentum. The Hamilton’s equations of motion are obtainable as coupled symplectic differential equations.  In this talk I shall show how optimized decision making (action sequences) can be obtained via a reinforcement learning problem wherein the agent interacts with the unknown environment to simultaneously learn a Hamiltonian surrogate and the optimal action sequences using Hamilton dynamics\, by invoking the Pontryagin Maximum Principle. We use optimal control theory to define an optimal control gradient flow\, which guides the reinforcement learning process of the agent to progressively optimize the Hamiltonian while simultaneously converging to the optimal action sequence. Extensions to stochastic Hamiltonians leading to stochastic action sequences and the free-energy principle shall also be discussed. This is joint work with  Taemin Heo\, Minh Nguyen. URL:https://ifds.info/event/optimized-decision-making-via-active-learning-of-stochastic-hamiltonians/ LOCATION:CSE (Allen) 403 CATEGORIES:MLOpt@UWash END:VEVENT END:VCALENDAR