Web navigation agents have made significant progress, yet current systems operate with no memory of past experiences -- leading to repeated mistakes and an inability to learn from previous interactions. We introduce Reflection-Augment Planning (ReAP), a web navigation system to leverage both successful and failed past experiences using self-reflections. Our method improves baseline results by 11 points overall and 29 points on previously failed tasks. These findings demonstrate that reflections can transfer to different web navigation tasks.

The development of effective explainability tools for Transformers is a crucial pursuit in deep learning research. One of the most promising approaches in this domain is Layer-wise Relevance Propagation (LRP), which propagates relevance scores backward through the network to the input space by redistributing activation values based on predefined rules. However, existing LRP-based methods for Transformer explainability entirely overlook a critical component of the Transformer architecture: its positional encoding (PE), resulting in violation of the conservation property, and the loss of an important and unique type of relevance, which is also associated with structural and positional features. To address this limitation, we reformulate the input space for Transformer explainability as a set of position-token pairs. This allows us to propose specialized theoretically-grounded LRP rules designed to propagate attributions across various positional encoding methods, including Rotary, Learnable, and Absolute PE. Extensive experiments with both fine-tuned classifiers and zero-shot foundation models, such as LLaMA 3, demonstrate that our method significantly outperforms the state-of-the-art in both vision and NLP explainability tasks. Our code is publicly available.

Recent advances in reasoning-enhanced Large Language Models such as OpenAI-o1/3 and DeepSeek-R1 have significantly improved performance on complex tasks. However, the quality and transparency of their internal reasoning processes remain underexplored. This work moves beyond the final-answer accuracy and investigates step-by-step reasoning in the medical and mathematical domains by explicitly decomposing the thinking trajectories into two parts: knowledge and reasoning. Specifically, we introduce a fine-grained evaluation framework that judges: (1) the correctness of knowledge used (measured by Knowledge Index (KI)) and (2) the quality of reasoning (measured by Information Gain (InfoGain)). Using this framework, we study R1-distilled and base Qwen models trained with supervised fine-tuning (SFT) and/or reinforcement learning (RL) in the medical and math domains. Three intriguing findings emerge: (1) The general reasoning abilities in R1-distilled models do not transfer effectively to the medical domain through either SFT or RL. (2) SFT raises final-answer accuracy in both domains, but often at the cost of reasoning quality: InfoGain drops by 38.9% on average compared with untrained models; In the medical domain, however, SFT remains crucial because domain knowledge is indispensable. (3) RL enhances medical reasoning by pruning inaccurate or irrelevant knowledge from reasoning paths, thereby improving both reasoning accuracy and knowledge correctness.

Effective decision making in partially observable environments requires compressing long interaction histories into informative representations. We introduce Descriptive History Representations (DHRs): sufficient statistics characterized by their capacity to answer relevant questions about past interactions and potential future outcomes. DHRs focus on capturing the information necessary to address task-relevant queries, providing a structured way to summarize a history for optimal control. We propose a multi-agent learning framework, involving representation, decision, and question-asking components, optimized using a joint objective that balances reward maximization with the representation's ability to answer informative questions. This yields representations that capture the salient historical details and predictive structures needed for effective decision making. We validate our approach on user modeling tasks with public movie and shopping datasets, generating interpretable textual user profiles which serve as sufficient statistics for predicting preference-driven behavior of users.

Inspired by the success of large language models (LLMs) in natural language processing, recent research has explored the building of time series foundation models and applied them to tasks such as forecasting, classification, and anomaly detection. However, their performances vary between different domains and tasks. In LLM-based approaches, test-time adaptation using example-based prompting has become common, owing to the high cost of retraining. In the context of anomaly detection, which is the focus of this study, providing normal examples from the target domain can also be effective. However, time series foundation models do not naturally acquire the ability to interpret or utilize examples or instructions, because the nature of time series data used during training does not encourage such capabilities. To address this limitation, we propose a retrieval augmented time series foundation model (RATFM), which enables pretrained time series foundation models to incorporate examples of test-time adaptation. We show that RATFM achieves a performance comparable to that of in-domain fine-tuning while avoiding domain-dependent fine-tuning. Experiments on the UCR Anomaly Archive, a multi-domain dataset including nine domains, confirms the effectiveness of the proposed approach.

Decomposing weight matrices into quantization and low-rank components ($\mathbf{W} \approx \mathbf{Q} + \mathbf{L}\mathbf{R}$) is a widely used technique for compressing large language models (LLMs). Existing joint optimization methods iteratively alternate between quantization and low-rank approximation. However, these methods tend to prioritize one component at the expense of the other, resulting in suboptimal decompositions that fail to leverage each component's unique strengths. In this work, we introduce Outlier-Driven Low-Rank Initialization (ODLRI), which assigns low-rank components the specific role of capturing activation-sensitive weights. This structured decomposition mitigates outliers' negative impact on quantization, enabling more effective balance between quantization and low-rank approximation. Experiments on Llama2 (7B, 13B, 70B), Llama3-8B, and Mistral-7B demonstrate that incorporating ODLRI into the joint optimization framework consistently reduces activation-aware error, minimizes quantization scale, and improves perplexity and zero-shot accuracy in low-bit settings.

Accurate evaluation of large language models (LLMs) is crucial for understanding their capabilities and guiding their development. However, current evaluations often inconsistently reflect the actual capacities of these models. In this paper, we demonstrate that one of many contributing factors to this \textit{evaluation crisis} is the oversight of unseen knowledge -- information encoded by LLMs but not directly observed or not yet observed during evaluations. We introduce KnowSum, a statistical framework designed to provide a more comprehensive assessment by quantifying the unseen knowledge for a class of evaluation tasks. KnowSum estimates the unobserved portion by extrapolating from the appearance frequencies of observed knowledge instances. We demonstrate the effectiveness and utility of KnowSum across three critical applications: estimating total knowledge, evaluating information retrieval effectiveness, and measuring output diversity. Our experiments reveal that a substantial volume of knowledge is omitted when relying solely on observed LLM performance. Importantly, KnowSum yields significantly different comparative rankings for several common LLMs based on their internal knowledge.

Enabling robots to accurately interpret and execute spoken language instructions is essential for effective human-robot collaboration. Traditional methods rely on speech recognition to transcribe speech into text, often discarding crucial prosodic cues needed for disambiguating intent. We propose a novel approach that directly leverages speech prosody to infer and resolve instruction intent. Predicted intents are integrated into large language models via in-context learning to disambiguate and select appropriate task plans. Additionally, we present the first ambiguous speech dataset for robotics, designed to advance research in speech disambiguation. Our method achieves 95.79% accuracy in detecting referent intents within an utterance and determines the intended task plan of ambiguous instructions with 71.96% accuracy, demonstrating its potential to significantly improve human-robot communication.

This paper presents a theoretical framework for addressing the challenges posed by generative artificial intelligence (AI) in higher education assessment through a machine-versus-machine approach. Large language models like GPT-4, Claude, and Llama increasingly demonstrate the ability to produce sophisticated academic content, traditional assessment methods face an existential threat, with surveys indicating 74-92% of students experimenting with these tools for academic purposes. Current responses, ranging from detection software to manual assessment redesign, show significant limitations: detection tools demonstrate bias against non-native English writers and can be easily circumvented, while manual frameworks rely heavily on subjective judgment and assume static AI capabilities. This paper introduces a dual strategy paradigm combining static analysis and dynamic testing to create a comprehensive theoretical framework for assessment vulnerability evaluation. The static analysis component comprises eight theoretically justified elements: specificity and contextualization, temporal relevance, process visibility requirements, personalization elements, resource accessibility, multimodal integration, ethical reasoning requirements, and collaborative elements. Each element addresses specific limitations in generative AI capabilities, creating barriers that distinguish authentic human learning from AI-generated simulation. The dynamic testing component provides a complementary approach through simulation-based vulnerability assessment, addressing limitations in pattern-based analysis. The paper presents a theoretical framework for vulnerability scoring, including the conceptual basis for quantitative assessment, weighting frameworks, and threshold determination theory.

Multimodal Continual Instruction Tuning (MCIT) aims to finetune Multimodal Large Language Models (MLLMs) to continually align with human intent across sequential tasks. Existing approaches often rely on the Mixture-of-Experts (MoE) LoRA framework to preserve previous instruction alignments. However, these methods are prone to Catastrophic Forgetting (CF), as they aggregate all LoRA blocks via simple summation, which compromises performance over time. In this paper, we identify a critical parameter inefficiency in the MoELoRA framework within the MCIT context. Based on this insight, we propose BranchLoRA, an asymmetric framework to enhance both efficiency and performance. To mitigate CF, we introduce a flexible tuning-freezing mechanism within BranchLoRA, enabling branches to specialize in intra-task knowledge while fostering inter-task collaboration. Moreover, we incrementally incorporate task-specific routers to ensure an optimal branch distribution over time, rather than favoring the most recent task. To streamline inference, we introduce a task selector that automatically routes test inputs to the appropriate router without requiring task identity. Extensive experiments on the latest MCIT benchmark demonstrate that BranchLoRA significantly outperforms MoELoRA and maintains its superiority across various MLLM sizes.