Searcharxiv⌕ Search

arXiv subjects

Hosein Hasanbeig

Publications and source records attributed to Hosein Hasanbeig.

15 recordsLinked to original sources

An Empirical Investigation of Robustness in Large Language Models under Tabular Distortions

Distortions are a natural consequence of tables collected from real-world sources, where scraping, extraction, or manual intervention can introduce semantic and structural inconsistencies. Humans can often identify and mentally correct such errors autonomously before reasoning over the data. In this work, we investigate whether Large Language Models (LLMs) possess a similar capability by evaluating them on 506 distorted table question answering (TQA) samples derived from canonical tables. Our findings reveal systematic weaknesses in current LLM pre-training paradigms, where models exhibit a strong inductive bias toward assuming tables are well-formed, often overlooking distortions or discarding inconsistent entries instead of repairing them. Even frontier models like GPT-5.4 and Opus-4.6 achieve only 68% accuracy under distortion. While distortion-aware prompting partially mitigates these failures, errors persist. Through lightweight finetuning, we demonstrate that models can learn to autonomously detect and repair human-recoverable distorted tables.

cs.AI↗

Ask Precisely, Assume Wisely: Interactive Evaluation of Ambiguity Resolution in Large Language Models

Ambiguity is a core challenge in human--AI interaction, where models must decide when to make assumptions and when to seek clarification. Existing approaches to ambiguity evaluation largely focus on coverage over multiple interpretations using static metrics, offering limited insight into how efficiently a model converges toward a single user-intended interpretation through interaction. We introduce a preference-grounded simulated user framework for controlled, multi-turn evaluation of ambiguity resolution, grounding simulator responses strictly in user disambiguation preferences to avoid hallucination and over-disclosure. Alongside this, we define three observable dissatisfaction signals --- incorrect assumptions, repeating clarifications, and irrelevant clarifications --- that characterize interaction quality beyond task accuracy. Our analysis reveals a clear trade-off: models that frequently seek clarification tend to over-clarify and prolong interactions, while assumption-driven models are more prone to misaligning with user intent. We demonstrate that system prompts optimized on these dissatisfaction signals achieve a Pareto-optimal balance across all three metrics, reducing overall error rates by up to 7.65%, underscoring the value of our framework in designing agents that ask precisely and assume wisely.

cs.AI↗

FlashScript: Compiling Recurring Agent Workflows into Reusable Neuro-Symbolic Programs

Many agent tasks recur with new data or small changes, yet agents regenerate their reasoning and actions on each run, producing different results. Distilling a prior run into a skill helps the agent avoid earlier mistakes, but it still re-runs the full process each time, so runs can vary and incur that cost. We compile a successful agent run into a reusable neuro-symbolic program (NSP) with a set of guidelines we call FlashScript. Actions and data transformations become code, instance-specific values become parameters, and decisions that need model judgment become LLM calls. We identify three layers where these calls belong and three techniques that replace most of them with code. The compilation does not remove nondeterminism; it localizes it, so the deterministic program and its neural calls can be reviewed and tested separately. Across 120 WorkBench analytics tasks with Claude Haiku 4.5, the NSPs were correct and made identical tool calls on every run, using about 70 times fewer input tokens and a third of the runtime of a skill-guided agent whose runs were inconsistent. In a composite version of the tasks, the skill-guided agent's accuracy fell to 23%, and its output inconsistency increased, while the NSPs remained correct and identical across all runs. The NSPs successfully generalized to all 108 unseen instances of the same templates. Our findings suggest that NSPs from agent trajectories can be considered cumulative software assets for end-users to test, maintain, and reuse in their recurring workflows.

cs.HC↗

RubricCraft: Where Solutions Diverge, Criteria Emerge

Evaluating language agents on open-ended, artifact-producing tasks is difficult: model-based graders can be inconsistent, while exhaustive programmatic checks are often impractical. Textual rubrics make evaluation criteria explicit, but writing them requires domain expertise. Rubrics generated from task descriptions alone can miss requirements and failure modes that become apparent in attempted solutions. We introduce RubricCraft, which treats rubric construction as the generation of natural-language specifications guided by candidate solutions. Given a task description and independently sampled solver trajectories, RubricCraft compares how each attempt satisfies or violates the task requirements and uses these differences to construct rubric criteria. It applies the criteria to the collected solutions and revises checks that overlook errors or penalize valid alternatives. The resulting rubrics capture partial correctness without requiring any single solution to serve as a complete reference. On SpreadsheetBench v2, against a prompt-only rubric written and applied by the same model, RubricCraft rubrics pass fewer wrong outputs (29.5% against 42.6%), fail none of the outputs the benchmark's own checker accepts, and rank outputs by correctness more often, while separating models as often. Repair on a single attempt leaves grading unchanged, but repair driven by attempts from three model families acts on different tasks, and an evidence audit judges most of those repairs improvements, many on tasks whose reference solution is itself wrong.

cs.SE↗

Whose turn is it? Context-guided model routing for stateful agents

Model routing can reduce the cost of LLM applications by assigning simpler requests to cheaper models without sacrificing the quality of the response. However, most routers treat requests as independent. Stateful agents do not operate under this assumption: a user turn may require multiple model and tool calls, its difficulty depends on the state produced by earlier turns, and changing models can sacrifice reuse of the conversational prompt cache. We study zero-shot, cost-aware routing at the user-turn boundary, systematically varying the routing rule, access to session history, candidate models that differ in size, reasoning effort, or model provider, and prompt-cache behavior. We evaluate on multi-turn versions of SheetCopilot and SpreadsheetBench 2, where persistent workbooks and cross-turn dependencies make isolated routing decisions especially brittle. At a matched offloading rate, giving the router session history improves task pass rate over the same router without history, with gains for every model pair; history-aware routing remains within one point of always using the expensive model. We also found that cost depends on execution dynamics rather than token prices alone: the cheaper model's per-task cost advantage narrows from 6.8x to 3.2x on longer sessions. A closed form cache analysis further shows that downward switches (expensive to cheap) recover their cache cost within two LLM calls, whereas upward switches do not break even under the evaluated conditions. Together, these results show that agent routing is a sequential decision problem: effective policies must account for what happened earlier in the session and for the asymmetric cost of switching models.

cs.AI↗

Distributions In, Distributions Out: The Case for Soft-Label Training

Supervised classifiers output a distribution over classes but are typically trained against a single label obtained by collapsing multiple annotators into a majority vote. On tasks where annotator disagreement reflects genuine ambiguity -- natural language inference, politeness, visually ambiguous categorization -- this collapse discards information and forces models to express uniform confidence on inputs where humans systematically disagree. We compare soft-label training, which uses the full annotation distribution as the target, against hard-label training across three datasets spanning vision and NLP (ChaosNLI, POPQUORN, CIFAR-10H). Soft-label training matches or exceeds hard-label accuracy on every dataset, reduces KL divergence to the annotator distribution by 32% on average (p < 10^-4), and produces predictions whose per-sample entropy correlates 61% more strongly with annotator entropy -- models trained on distributions are uncertain precisely where humans are. We argue these benefits follow from a basic observation: when annotators legitimately disagree, the annotation distribution is the correct learning target, not a noisy estimate of it.

cs.LG↗

ConDABench: Interactive Evaluation of Language Models for Data Analysis

Real-world data analysis tasks often come with under-specified goals and unclean data. User interaction is necessary to understand and disambiguate a user's intent, and hence, essential to solving these complex tasks. Existing benchmarks for evaluating LLMs on data analysis tasks do not capture these complexities or provide first-class support for interactivity. We introduce ConDABench, a framework for generating conversational data analysis (ConDA) benchmarks and evaluating external tools on the generated benchmarks. \bench consists of (a) a multi-agent workflow for generating realistic benchmarks from articles describing insights gained from public datasets, (b) 1,420 ConDA problems generated using this workflow, and (c) an evaluation harness that, for the first time, makes it possible to systematically evaluate conversational data analysis tools on the generated ConDA problems. Evaluation of state-of-the-art LLMs on the benchmarks reveals that while the new generation of models are better at solving more instances, they are not necessarily better at solving tasks that require sustained, long-form engagement. ConDABench is an avenue for model builders to measure progress towards truly collaborative models that can complete complex interactive tasks.

cs.CL↗

Mission-driven Exploration for Accelerated Deep Reinforcement Learning with Temporal Logic Task Specifications

This paper addresses the problem of designing control policies for agents with unknown stochastic dynamics and control objectives specified using Linear Temporal Logic (LTL). Recent Deep Reinforcement Learning (DRL) algorithms have aimed to compute policies that maximize the satisfaction probability of LTL formulas, but they often suffer from slow learning performance. To address this, we introduce a novel Deep Q-learning algorithm that significantly improves learning speed. The enhanced sample efficiency stems from a mission-driven exploration strategy that prioritizes exploration towards directions likely to contribute to mission success. Identifying these directions relies on an automaton representation of the LTL task as well as a learned neural network that partially models the agent-environment interaction. We provide comparative experiments demonstrating the efficiency of our algorithm on robot navigation tasks in unseen environments.

cs.RO↗

Progressive Safeguards for Safe and Model-Agnostic Reinforcement Learning

In this paper we propose a formal, model-agnostic meta-learning framework for safe reinforcement learning. Our framework is inspired by how parents safeguard their children across a progression of increasingly riskier tasks, imparting a sense of safety that is carried over from task to task. We model this as a meta-learning process where each task is synchronized with a safeguard that monitors safety and provides a reward signal to the agent. The safeguard is implemented as a finite-state machine based on a safety specification; the reward signal is formally shaped around this specification. The safety specification and its corresponding safeguard can be arbitrarily complex and non-Markovian, which adds flexibility to the training process and explainability to the learned policy. The design of the safeguard is manual but it is high-level and model-agnostic, which gives rise to an end-to-end safe learning approach with wide applicability, from pixel-level game control to language model fine-tuning. Starting from a given set of safety specifications (tasks), we train a model such that it can adapt to new specifications using only a small number of training samples. This is made possible by our method for efficiently transferring safety bias between tasks, which effectively minimizes the number of safety violations. We evaluate our framework in a Minecraft-inspired Gridworld, a VizDoom game environment, and an LLM fine-tuning application. Agents trained with our approach achieve near-minimal safety violations, while baselines are shown to underperform.

cs.LG↗

Decoding In-Context Learning: Neuroscience-inspired Analysis of Representations in Large Language Models

Large language models (LLMs) exhibit remarkable performance improvement through in-context learning (ICL) by leveraging task-specific examples in the input. However, the mechanisms behind this improvement remain elusive. In this work, we investigate how LLM embeddings and attention representations change following in-context-learning, and how these changes mediate improvement in behavior. We employ neuroscience-inspired techniques such as representational similarity analysis (RSA) and propose novel methods for parameterized probing and measuring ratio of attention to relevant vs. irrelevant information in Llama-2 70B and Vicuna 13B. We designed two tasks with a priori relationships among their conditions: linear regression and reading comprehension. We formed hypotheses about expected similarities in task representations and measured hypothesis alignment of LLM representations before and after ICL as well as changes in attention. Our analyses revealed a meaningful correlation between improvements in behavior after ICL and changes in both embeddings and attention weights across LLM layers. This empirical framework empowers a nuanced understanding of how latent representations shape LLM behavior, offering valuable tools and insights for future research and practical applications.

cs.CL↗

Safeguarded Progress in Reinforcement Learning: Safe Bayesian Exploration for Control Policy Synthesis

This paper addresses the problem of maintaining safety during training in Reinforcement Learning (RL), such that the safety constraint violations are bounded at any point during learning. In a variety of RL applications the safety of the agent is particularly important, e.g. autonomous platforms or robots that work in proximity of humans. As enforcing safety during training might severely limit the agent's exploration, we propose here a new architecture that handles the trade-off between efficient progress and safety during exploration. As the exploration progresses, we update via Bayesian inference Dirichlet-Categorical models of the transition probabilities of the Markov decision process that describes the environment dynamics. This paper proposes a way to approximate moments of belief about the risk associated to the action selection policy. We construct those approximations, and prove the convergence results. We propose a novel method for leveraging the expectation approximations to derive an approximate bound on the confidence that the risk is below a certain level. This approach can be easily interleaved with RL and we present experimental results to showcase the performance of the overall architecture.

cs.LG↗

ALLURE: Auditing and Improving LLM-based Evaluation of Text using Iterative In-Context-Learning

From grading papers to summarizing medical documents, large language models (LLMs) are evermore used for evaluation of text generated by humans and AI alike. However, despite their extensive utility, LLMs exhibit distinct failure modes, necessitating a thorough audit and improvement of their text evaluation capabilities. Here we introduce ALLURE, a systematic approach to Auditing Large Language Models Understanding and Reasoning Errors. ALLURE involves comparing LLM-generated evaluations with annotated data, and iteratively incorporating instances of significant deviation into the evaluator, which leverages in-context learning (ICL) to enhance and improve robust evaluation of text by LLMs. Through this iterative process, we refine the performance of the evaluator LLM, ultimately reducing reliance on human annotators in the evaluation process. We anticipate ALLURE to serve diverse applications of LLMs in various domains related to evaluation of textual data, such as medical summarization, education, and and productivity.

cs.CL↗

Evaluating Cognitive Maps and Planning in Large Language Models with CogEval

Recently an influx of studies claim emergent cognitive abilities in large language models (LLMs). Yet, most rely on anecdotes, overlook contamination of training sets, or lack systematic Evaluation involving multiple tasks, control conditions, multiple iterations, and statistical robustness tests. Here we make two major contributions. First, we propose CogEval, a cognitive science-inspired protocol for the systematic evaluation of cognitive capacities in Large Language Models. The CogEval protocol can be followed for the evaluation of various abilities. Second, here we follow CogEval to systematically evaluate cognitive maps and planning ability across eight LLMs (OpenAI GPT-4, GPT-3.5-turbo-175B, davinci-003-175B, Google Bard, Cohere-xlarge-52.4B, Anthropic Claude-1-52B, LLaMA-13B, and Alpaca-7B). We base our task prompts on human experiments, which offer both established construct validity for evaluating planning, and are absent from LLM training sets. We find that, while LLMs show apparent competence in a few planning tasks with simpler structures, systematic evaluation reveals striking failure modes in planning tasks, including hallucinations of invalid trajectories and getting trapped in loops. These findings do not support the idea of emergent out-of-the-box planning ability in LLMs. This could be because LLMs do not understand the latent relational structures underlying planning problems, known as cognitive maps, and fail at unrolling goal-directed trajectories based on the underlying structure. Implications for application and future directions are discussed.

cs.AI↗

Certified Reinforcement Learning with Logic Guidance

Reinforcement Learning (RL) is a widely employed machine learning architecture that has been applied to a variety of control problems. However, applications in safety-critical domains require a systematic and formal approach to specifying requirements as tasks or goals. We propose a model-free RL algorithm that enables the use of Linear Temporal Logic (LTL) to formulate a goal for unknown continuous-state/action Markov Decision Processes (MDPs). The given LTL property is translated into a Limit-Deterministic Generalised Buchi Automaton (LDGBA), which is then used to shape a synchronous reward function on-the-fly. Under certain assumptions, the algorithm is guaranteed to synthesise a control policy whose traces satisfy the LTL specification with maximal probability.

cs.LG↗

LCRL: Certified Policy Synthesis via Logically-Constrained Reinforcement Learning

LCRL is a software tool that implements model-free Reinforcement Learning (RL) algorithms over unknown Markov Decision Processes (MDPs), synthesising policies that satisfy a given linear temporal specification with maximal probability. LCRL leverages partially deterministic finite-state machines known as Limit Deterministic Buchi Automata (LDBA) to express a given linear temporal specification. A reward function for the RL algorithm is shaped on-the-fly, based on the structure of the LDBA. Theoretical guarantees under proper assumptions ensure the convergence of the RL algorithm to an optimal policy that maximises the satisfaction probability. We present case studies to demonstrate the applicability, ease of use, scalability, and performance of LCRL. Owing to the LDBA-guided exploration and LCRL model-free architecture, we observe robust performance, which also scales well when compared to standard RL approaches (whenever applicable to LTL specifications). Full instructions on how to execute all the case studies in this paper are provided on a GitHub page that accompanies the LCRL distribution www.github.com/grockious/lcrl.

cs.LG↗