Portfolio item number 1
Short description of portfolio item number 1
Short description of portfolio item number 1
Short description of portfolio item number 2
Published in Advances in Neural Information Processing Systems (NeurIPS), 2018
We introduced the problem of lifelong learning from demonstrations, and created an efficient lifelong inverse reinforcement learning (ELIRL) algorithm.
Download here
Published in 3rd Conversational AI Workshop at NeurIPS, 2019
We developed an architecture for learning multi-domain task oriented dialog policies, based on the notion of action embeddings, which capture domain agnostic representations of how to respond to user’s queries.
Download here
Published in Advances in Neural Information Processing Systems (NeurIPS), 2019
We introduced the notion of performance gap as a label-dependent notion of domain discrepancy, and developed an boosting-based algorithm, gapBoost, that exploits the insights from gap minimization.
Download here
Published in Advances in Neural Information Processing Systems (NeurIPS), 2020
We introduced an algorithm for directly optimizing factored policies via policy gradients in a lifelong learning setting, and showed theoretically and empirically that our approach avoids catastrophic forgetting.
Download here
Published in International Conference on Learning Representations (ICLR), 2021
We study the question of how to learn compositional parameterized structures from an empirical standpoint, and propose a general-purpose framework that can learn with various forms of knowledge representations and base algorithms.
Download here
Published in University of Pennsylvania, 2022
This dissertation presents an in-depth treatment of the problems of lifelong or continual learning and compositional knowledge representations, which had so far been studied separately.
Download here
Published in International Conference on Learning Representations (ICLR), 2022
We explore the problem of lifelong RL of functionally compositional knowledge, and develop an algorithm that demonstrates zero-shot and forward transfer, avoidance of forgetting, and backward transfer in discrete 2-D and robotic manipulation domains.
Download here
Published in Conference on Lifelong Learning Agents (CoLLAs), 2022
We introduce CompoSuite, a robotic manipulation benchmark with hundreds of tasks for evaluating the functional compositionality of RL algorithms.
Download here
Published in Conference on Lifelong Learning Agents (CoLLAs), 2022
We introduce SHELS, a combined framework that supports class-incremental continual learning without the specification of explicit class boundaries by performing out-of-distribution detection.
Download here
Published in CoRL Workshop on Pre-training Robot Learning, 2022
We propose and release a variety of data sets for compositional off-line RL on CompoSuite
Download here
Published in Journal on Machine Learning Research (JMLR), 2023
We extended the notion of performance gap for measuring domain discrepancy (NeurIPS-19) to a variety of transfer and multi-task learning settings, and introduced two new algorithms based on this notion for improving transfer and multi-task learning performance.
Download here
Published in Neural Networks, 2023
We propose a unified approach to assess the performance of lifelong learning approaches, agnostic to the specific domain or technique used for learning.
Download here
Published in Transactions on Machine Learning Research (TMLR), 2023
We survey the mostly disjoint fields of lifelong or continual learning and compositional learning, and draw connections between them.
Download here
Published in Conference on Robot Learning (CoRL), 2023
We formulate a realistic variant of the problem of lifelong learning for TAMP, and devise a mixture of generative models for generating samples for efficient planning.
Download here
Published in CoLLAs Workshop Track, 2023
We introduce a simple cross-validation-based procedure to automatically tune the threshold for continual novelty detection using SHELS.
Download here
Published in Transactions on Machine Learning Research (TMLR), 2024
We propose a test-time augmentation approach that leverages LLMs to transform out-of-distribution data into in-distribution to improve robustness of task-specific natural language classifiers.
Download here
Published in Reinforcement Learning Conference (RLC), 2024
We release four datasets of simulated robotic manipulation trajectories for offline compositional reinforcement learning.
Download here