David H. Yang, Yuxuan Zhu, Mohammad Mohammadi Amiri, Keerthiram Murugesan, Tejaswini Pedapati, Subhajit Chaudhury, Pin-Yu Chen. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Zheyuan Zhang, Kaiwen Shi, Zhengqing Yuan, Zehong Wang, Tianyi Ma, Keerthiram Murugesan, Vincent Galassi, Chuxu Zhang, Yanfang Ye. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Zheyuan Zhang, Yiyang Li, Nhi Ha Lan Le, Zehong Wang, Tianyi Ma, Vincent Galassi, Keerthiram Murugesan, Nuno Moniz, Werner Geyer, Nitesh V Chawla, Chuxu Zhang, Yanfang Ye. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Text-based reinforcement learning involves an agent interacting with a fictional environment using observed text and admissible actions in natural language to complete a task. Previous works have shown that agents can succeed in text-based interactive environments even in the complete absence of semantic understanding or other linguistic capabilities. The success of these agents in playing such games suggests that semantic understanding may not be important for the task. This raises an important question about the benefits of LMs in guiding the agents through the game states. In this work, we show that rich semantic understanding leads to efficient training of text-based RL agents. Moreover, we describe the occurrence of semantic degeneration as a consequence of inappropriate fine-tuning of language models in text-based reinforcement learning (TBRL). Specifically, we describe the shift in the semantic representation of words in the LM, as well as how it affects the performance of the agent in tasks that are semantically similar to the training games. These results may help develop better strategies to fine-tune agents in text-based RL scenarios.

The prevalence of unhealthy eating habits has become a growing concern in the United States. However, popular food recommendation platforms, such as Yelp, tend to prioritize users' dietary preferences over the healthiness of their choices. While some efforts have focused on developing health-aware food recommendation systems, personalization based on specific health conditions remains underexplored. Additionally, the lack of interpretability in these systems prevents users from evaluating the reliability of recommendations, limiting their practical adoption. To address these issues, we introduce two large-scale personalized health-aware food recommendation benchmarks at the first attempt. Building on this, we propose a novel framework called the Multi-Objective Personalized Interpretable Health-aware Food Recommendation System (MOPI-HFRS). This system generates food recommendations by jointly optimizing three objectives: user preference, personalized healthiness, and nutritional diversity. It also incorporates a reasoning module enhanced by large language models (LLMs) to provide interpretable recommendations that promote healthy dietary knowledge. The framework integrates descriptive features and health data using two structure learning and pooling modules within a graph learning framework. Pareto optimization is applied to balance the multi-faceted objectives. To further enhance healthy dietary knowledge, the system leverages LLMs by infusing knowledge from the recommendation model, generating meaningful interpretations for the recommendations. Extensive experiments on the proposed benchmarks demonstrate that MOPI-HFRS outperforms state-of-the-art methods by delivering diverse, healthy food recommendations alongside reliable explanations.

The exponential growth of data-driven systems and AI technologies has intensified the demand for high-quality web-sourced datasets. While existing datasets have proven valuable, conventional web data collection approaches face significant limitations in terms of human effort and scalability. Current data collecting solutions fall into two categories: wrapper-based methods that struggle with adaptability and reproducibility, and large language model (LLM)-based approaches that incur substantial computational and financial costs. To address these challenges, we propose AutoData, a novel multi-agent system for Automated web Data collection, that requires minimal human intervention, i.e., only necessitating a natural language instruction specifying the desired dataset. In addition, AutoData is designed for a robust multi-agent architecture, featuring a novel oriented message hypergraph coordinated by a central task manager, to efficiently organize agents across research and development squads. Besides, we introduce a novel hypergraph cache system to advance the multi-agent collaboration process that enables efficient automated data collection and mitigates the token cost issues prevalent in existing LLM-based systems. Moreover, we introduce Instruct2DS, a new benchmark dataset supporting live data collection from web sources across three domains: academic, finance, and sports. Comprehensive evaluations over Instruct2DS and three existing benchmark datasets demonstrate AutoData's superior performance compared to baseline methods. Case studies on challenging tasks such as picture book collection and paper extraction from surveys further validate its applicability.

Manual evaluation of Large Language Model (LLM) applications at scale presents significant resource challenges, making LLM-as-Judge (LaaJ) an attractive alternative. This study examines the reliability of LaaJ evaluation within CT-Bench, a benchmark for assessing LLMs’ capabilities in recommending clinical trial baseline features. LaaJ-alpha, our GPT-4o based prototype, semantically matches LLM-recommended features against reference features from clinical trials, accounting for semantic equivalence (e.g., ‘BMI’ and ‘Body Mass Index’). The system generates matched pairs and unmatched features from both sources to calculate precision, recall, and F1 scores. Laaj-alpha evaluates baseline feature recommendations across CTBench CT-Pub (100 trials) and CT-Repo (1,690 trials) for comparing results for GPT-4o and Llama-3-70B-Instruct under zero-shot and three-shot settings. Coherence checking revealed hallucinations in LaaJ-alpha’s evaluation, necessitating a post-processing correction step that yielded lower but more accurate performance metrics. Three different types of hallucination were observed. The hallucination rate provides a quantifiable coherence metric that can be systematically used to improve LaaJ reliability. Our findings underscore the challenges in developing reliable LLM evaluation methods in healthcare applications and demonstrate a potential framework for improving LaaJ systems.

Kinjal Basu, Keerthiram Murugesan, Subhajit Chaudhury, Murray Campbell, Kartik Talamadupula, Tim Klinger. Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

We present ALT (ALignment with Textual feedback), an approach that aligns language models with user preferences expressed in text.We argue that text offers greater expressiveness, enabling users to provide richer feedback than simple comparative preferences, leading to more efficient and effective alignment.ALT aligns the model by conditioning its generations on the textual feedback.Our method relies solely on language modeling techniques and requires minimal hyper-parameter tuning while retaining the main benefits of RL-based alignment algorithms.We demonstrate the efficacy and efficiency of textual feedback across different tasks, including toxicity reduction, summarization, and dialogue response generation.Notably, ALT outperforms PPO in toxicity reduction and matches its performance on summarization with only 20% of the samples.We also explore using ALT with feedback from an existing LLM, examining constrained and unconstrained feedback.Additionally, we outline future directions to align models with natural language feedback.1

Automated Planning and Scheduling is among the growing areas in Artificial Intelligence (AI) where mention of LLMs has gained popularity. Based on a comprehensive review of 126 papers, this paper investigates eight categories based on the unique applications of LLMs in addressing various aspects of planning problems: language translation, plan generation, model construction, multi-agent planning, interactive planning, heuristics optimization, tool integration, and brain-inspired planning. For each category, we articulate the issues considered and existing gaps. A critical insight resulting from our review is that the true potential of LLMs unfolds when they are integrated with traditional symbolic planners, pointing towards a promising neuro-symbolic approach. This approach effectively combines the generative aspects of LLMs with the precision of classical planning methods. By synthesizing insights from existing literature, we underline the potential of this integration to address complex planning challenges. Our goal is to encourage the ICAPS community to recognize the complementary strengths of LLMs and symbolic planners, advocating for a direction in automated planning that leverages these synergistic capabilities to develop more advanced and intelligent planning systems. We aim to keep the categorization of papers updated on https://ai4society.github.io/LLM-Planning-Viz/, a collaborative resource that allows researchers to contribute and add new literature to the categorization.

We introduce VisualHandicaps, a novel benchmark environment for the systematic analysis of interactive text-based reinforcement learning (TBRL) agents by providing visual handicaps. Unlike previous TBRL environments, which focus on providing additional textual information to measure agent understanding of sequential natural language information, VisualHandicaps seeks to improve the generalization ability of RL agents using varying details of maps and textual information, allowing for the study and demonstration of robust planning and self-localization. We provide automatically generated variations and difficulty levels in our environment and show that an agent using our systematic visual handicaps along with textual observation generally outperforms previous methods (that use only textual handicaps) in terms of success rate and the number of steps required to reach the goal. We also provide a detailed analysis of each handicap, which we believe to be important findings for driving future improvements in RL agents on text-based applications.

Many machine learning problems today have multiple objective functions, which are often tackled by the multi-objective learning (MOL) framework. Albeit many encouraging results are obtained by MOL algorithms, a recent theoretical study [1] revealed that these gradient-based MOL methods (e.g., MGDA, CAGrad) all reflect an inherent trade-off between optimization convergence speeds and conflict-avoidance abilities. To this end, we develop an improved stochastic variance-reduced multi-objective gradient correction method for MOL, achieving the ${\mathcal{O}}\left({{\varepsilon ^{ - 1.5}}}\right)$ sample complexity. In addition, our proposed method simultaneously improves the theoretical guarantees for conflict avoidance and convergence rate compared to prior stochastic gradient-based MOL methods in the non-convex setting. We further validate the effectiveness of the proposed method empirically using popular multi-task learning (MTL) benchmarks.

This paper studies the transfer reinforcement learning (RL) problem where multiple RL problems have different reward functions but share the same underlying transition dynamics. In this setting, the Q-function of each RL problem (task) can be decomposed into a successor feature (SF) and a reward mapping: the former characterizes the transition dynamics, and the latter characterizes the task-specific reward function. This Q-function decomposition, coupled with a policy improvement operator known as generalized policy improvement (GPI), reduces the sample complexity of finding the optimal Q-function, and thus the SF & GPI framework exhibits promising empirical performance compared to traditional RL methods like Q-learning. However, its theoretical foundations remain largely unestablished, especially when learning the successor features using deep neural networks (SF-DQN). This paper studies the provable knowledge transfer using SFs-DQN in transfer RL problems. We establish the first convergence analysis with provable generalization guarantees for SF-DQN with GPI. The theory reveals that SF-DQN with GPI outperforms conventional RL approaches, such as deep Q-network, in terms of both faster convergence rate and better generalization. Numerical experiments on real and synthetic RL tasks support the superior performance of SF-DQN & GPI, aligning with our theoretical findings.

Subhajit Chaudhury, Sarathkrishna Swaminathan, Daiki Kimura, Prithviraj Sen, Keerthiram Murugesan, Rosario Uceda-Sosa, Michiaki Tatsubori, Achille Fokoue, Pavan Kapanipathi, Asim Munawar, Alexander Gray. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Event sequences are widely available across application domains and there is a long history of models for representing and analyzing such datasets. Summary Markov models are a recent addition to the literature that help identify the subset of event types that influence event types of interest to a user. In this paper, we introduce logical summary Markov models, which are a family of models for event sequences that enable interpretable predictions through logical rules that relate historical predicates to the probability of observing an event type at any arbitrary position in the sequence. We illustrate their connection to prior parametric summary Markov models as well as probabilistic logic programs, and propose new models from this family along with efficient greedy search algorithms for learning them from data. The proposed models outperform relevant baselines on most datasets in an empirical investigation on a probabilistic prediction task. We also compare the number of influencers that various logical summary Markov models learn on real-world datasets, and conduct a brief exploratory qualitative study to gauge the promise of such symbolic models around guiding large language models for predicting societal events.

Plansformer is a novel tool that utilizes a fine-tuned language model based on transformer architecture to generate symbolic plans. Transformers are a type of neural network architecture that have been shown to be highly effective in a range of natural language processing tasks. Unlike traditional planning systems that use heuristic-based search strategies, Plansformer is fine-tuned on specific classical planning domains to generate high-quality plans that are both fluent and feasible. Plansformer takes the domain and problem files as input (in PDDL) and outputs a sequence of actions that can be executed to solve the problem. We demonstrate the effectiveness of Plansformer on a variety of benchmark problems and provide both qualitative and quantitative results obtained during our evaluation, including its limitations. Plansformer has the potential to significantly improve the efficiency and effectiveness of planning in various domains, from logistics and scheduling to natural language processing and human-computer interaction. In addition, we provide public access to Plansformer via a website as well as an API endpoint; this enables other researchers to utilize our tool for planning and execution. The demo video is available at https://youtu.be/_1rlctCGsrk

Text-based games (TBGs) have become a popular proving ground for the demonstration of learning-based agents that make decisions in quasi real-world settings. The crux of the problem for a reinforcement learning agent in such TBGs is identifying the objects in the world, and those objects' relations with that world. While the recent use of text-based resources for increasing an agent's knowledge and improving its generalization have shown promise, we posit in this paper that there is much yet to be learned from visual representations of these same worlds. Specifically, we propose to retrieve images that represent specific instances of text observations from the world and train our agents on such images. This improves the agent's overall understanding of the game scene and objects' relationships to the world around them, and the variety of visual representations on offer allow the agent to generate a better generalization of a relationship. We show that incorporating such images improves the performance of agents in various TBG settings.

Subhajit Chaudhury, Sarathkrishna Swaminathan, Chulaka Gunasekara, Maxwell Crouse, Srinivas Ravishankar, Daiki Kimura, Keerthiram Murugesan, Ramón Fernandez Astudillo, Tahira Naseem, Pavan Kapanipathi, Alexander Gray. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. 2022.

This paper proposes a research direction to advance AI which draws inspiration from cognitive theories of human decision making. The premise is that if we gain insights about the causes of some human capabilities that are still lacking in AI (for instance, adaptability, generalizability, common sense, and causal reasoning), we may obtain similar capabilities in an AI system by embedding these causal components. We hope that the high-level description of our vision included in this paper, as well as the several research questions that we propose to consider, can stimulate the AI research community to define, try and evaluate new methodologies, frameworks, and evaluation metrics, in the spirit of achieving a better understanding of both human and machine intelligence.

Text-based games have emerged as an important test-bed for Reinforcement Learning (RL) research, requiring RL agents to combine grounded language understanding with sequential decision making. In this paper, we examine the problem of infusing RL agents with commonsense knowledge. Such knowledge would allow agents to efficiently act in the world by pruning out implausible actions, and to perform look-ahead planning to determine how current actions might affect future world states. We design a new text-based gaming environment called TextWorld Commonsense (TWC) for training and evaluating RL agents with a specific kind of commonsense knowledge about objects, their attributes, and affordances. We also introduce several baseline RL agents which track the sequential context and dynamically retrieve the relevant commonsense knowledge from ConceptNet. We show that agents which incorporate commonsense knowledge in TWC perform better, while acting more efficiently. We conduct user-studies to estimate human performance on TWC and show that there is ample room for future improvement.

This paper introduces self-paced task selection to multitask learning, where instances from more closely related tasks are selected in a progression of easier-to-harder tasks, to emulate an effective human education strategy, but applied to multitask machine learning. We develop the mathematical foundation for the approach based on iterative selection of the most appropriate task, learning the task parameters, and updating the shared knowledge, optimizing a new bi-convex loss function. This proposed method applies quite generally, including to multitask feature learning, multitask learning with alternating structure optimization, etc. Results show that in each of the above formulations self-paced (easier-to-harder) task selection outperforms the baseline version of these methods in all the experiments.

This paper addresses the challenge of learning from peers in an online multitask setting. Instead of always requesting a label from a human oracle, the proposed method first determines if the learner for each task can acquire that label with sufficient confidence from its peers either as a task-similarity weighted sum, or from the single most similar task. If so, it saves the oracle query for later use in more difficult cases, and if not it queries the human oracle. The paper develops the new algorithm to exhibit this behavior and proves a theoretical mistake bound for the method compared to the best linear predictor in hindsight. Experiments over three multitask learning benchmark datasets show clearly superior performance over baselines such as assuming task independence, learning only from the oracle and not learning from peer tasks.

This paper presents a novel multitask multiple-kernel learning framework that efficiently learns the kernel weights leveraging the relationship across multiple tasks. The idea is to automatically infer this task relationship in the RKHS space corresponding to the given base kernels. The problem is formulated as a regularization-based approach called Multi-Task Multiple Kernel Relationship Learning (MK-MTRL), which models the task relationship matrix from the weights learned from latent feature spaces of task-specific base kernels. Unlike in previous work, the proposed formulation allows one to incorporate prior knowledge for simultaneously learning several related task. We propose an alternating minimization algorithm to learn the model parameters, kernel weights and task relationship matrix. In order to tackle large-scale problems, we further propose a two-stage MK-MTRL online learning algorithm and show that it significantly reduces the computational time, and also achieves performance comparable to that of the joint learning framework. Experimental results on benchmark datasets show that the proposed formulations outperform several state-of-the-art multitask learning methods.

This paper addresses the challenge of jointly learning both the per-task model parameters and the inter-task relationships in a multi-task online learning setting. The proposed algorithm features probabilistic interpretation, efficient updating rules and flexible modulation on whether learners focus on their specific task or on jointly address all tasks. The paper also proves a sub-linear regret bound as compared to the best linear predictor in hindsight. Experiments over three multi-task learning benchmark datasets show advantageous performance of the proposed approach over several state-of-the-art online multi-task learning baselines.