Adaptive exercise recommendation (ER) aims to choose the next activity that matches a learner's evolving Zone of Proximal Development (ZPD). We present KUL-Rec, a biologically inspired ER system that couples a fast Hebbian memory with slow replay-based consolidation to enable continual, few-shot personalization from sparse interactions. The model operates in an embedding space, allowing a single architecture to handle both tabular knowledge-tracing logs and open-ended short-answer text. We align evaluation with tutoring needs using bidirectional ranking and rank-sensitive metrics (nDCG, Recall@K). Across ten public datasets, KUL-Rec improves macro nDCG (0.316 vs. 0.265 for the strongest baseline) and Recall@10 (0.305 vs. 0.211), while achieving low inference latency and an $\approx99$\% reduction in peak GPU memory relative to a competitive graph-based model. In a 13-week graduate course, KUL-Rec personalized weekly short-answer quizzes generated by a retrieval-augmented pipeline and the personalized quizzes were associated with lower perceived difficulty and higher helpfulness (p < .05). An embedding robustness audit highlights that encoder choice affects semantic alignment, motivating routine audits when deploying open-response assessment. Together, these results indicate that Hebbian replay with bounded consolidation offers a practical path to real-time, interpretable ER that scales across data modalities and classroom settings.

Local-global attention models have recently emerged as compelling alternatives to standard Transformers, promising improvements in both training and inference efficiency. However, the crucial choice of window size presents a Pareto tradeoff: larger windows maintain performance akin to full attention but offer minimal efficiency gains in short-context scenarios, while smaller windows can lead to performance degradation. Current models, such as Gemma2 and Mistral, adopt conservative window sizes (e.g., 4096 out of an 8192 pretraining length) to preserve performance. This work investigates strategies to shift this Pareto frontier, enabling local-global models to achieve efficiency gains even in short-context regimes. Our core motivation is to address the intrinsic limitation of local attention -- its complete disregard for tokens outside the defined window. We explore RATTENTION, a variant of local attention integrated with a specialized linear attention mechanism designed to capture information from these out-of-window tokens. Pretraining experiments at the 3B and 12B scales demonstrate that RATTENTION achieves a superior Pareto tradeoff between performance and efficiency. As a sweet spot, RATTENTION with a window size of just 512 consistently matches the performance of full-attention models across diverse settings. Furthermore, the recurrent nature inherent in the linear attention component of RATTENTION contributes to enhanced long-context performance, as validated on the RULER benchmark. Crucially, these improvements do not compromise training efficiency; thanks to a specialized kernel implementation and the reduced window size, RATTENTION maintains training speeds comparable to existing state-of-the-art approaches. We open-sourced our Pallas kernels along with model codes to facilitate further research effort.

Accurate intent classification is critical for efficient routing in customer service, ensuring customers are connected with the most suitable agents while reducing handling times and operational costs. However, as companies expand their product lines, intent classification faces scalability challenges due to the increasing number of intents and variations in taxonomy across different verticals. In this paper, we introduce REIC, a Retrieval-augmented generation Enhanced Intent Classification approach, which addresses these challenges effectively. REIC leverages retrieval-augmented generation (RAG) to dynamically incorporate relevant knowledge, enabling precise classification without the need for frequent retraining. Through extensive experiments on real-world datasets, we demonstrate that REIC outperforms traditional fine-tuning, zero-shot, and few-shot methods in large-scale customer service settings. Our results highlight its effectiveness in both in-domain and out-of-domain scenarios, demonstrating its potential for real-world deployment in adaptive and large-scale intent classification systems.

Understanding functional representations within higher visual cortex is a fundamental question in computational neuroscience. While artificial neural networks pretrained on large-scale datasets exhibit striking representational alignment with human neural responses, learning image-computable models of visual cortex relies on individual-level, large-scale fMRI datasets. The necessity for expensive, time-intensive, and often impractical data acquisition limits the generalizability of encoders to new subjects and stimuli. BraInCoRL uses in-context learning to predict voxelwise neural responses from few-shot examples without any additional finetuning for novel subjects and stimuli. We leverage a transformer architecture that can flexibly condition on a variable number of in-context image stimuli, learning an inductive bias over multiple subjects. During training, we explicitly optimize the model for in-context learning. By jointly conditioning on image features and voxel activations, our model learns to directly generate better performing voxelwise models of higher visual cortex. We demonstrate that BraInCoRL consistently outperforms existing voxelwise encoder designs in a low-data regime when evaluated on entirely novel images, while also exhibiting strong test-time scaling behavior. The model also generalizes to an entirely new visual fMRI dataset, which uses different subjects and fMRI data acquisition parameters. Further, BraInCoRL facilitates better interpretability of neural signals in higher visual cortex by attending to semantically relevant stimuli. Finally, we show that our framework enables interpretable mappings from natural language queries to voxel selectivity.

Visual document retrieval requires understanding heterogeneous and multi-modal content to satisfy information needs. Recent advances use screenshot-based document encoding with fine-grained late interaction, significantly improving retrieval performance. However, retrievers are still trained with coarse global relevance labels, without revealing which regions support the match. As a result, retrievers tend to rely on surface-level cues and struggle to capture implicit semantic connections, hindering their ability to handle non-extractive queries. To alleviate this problem, we propose a \textbf{A}ttention-\textbf{G}rounded \textbf{RE}triever \textbf{E}nhancement (AGREE) framework. AGREE leverages cross-modal attention from multimodal large language models as proxy local supervision to guide the identification of relevant document regions. During training, AGREE combines local signals with the global signals to jointly optimize the retriever, enabling it to learn not only whether documents match, but also which content drives relevance. Experiments on the challenging ViDoRe V2 benchmark show that AGREE significantly outperforms the global-supervision-only baseline. Quantitative and qualitative analyses further demonstrate that AGREE promotes deeper alignment between query terms and document regions, moving beyond surface-level matching toward more accurate and interpretable retrieval. Our code is available at: https://anonymous.4open.science/r/AGREE-2025.

We propose a Kardashev-inspired yet operational Autonomous AI (AAI) Scale that measures the progression from fixed robotic process automation (AAI-0) to full artificial general intelligence (AAI-4) and beyond. Unlike narrative ladders, our scale is multi-axis and testable. We define ten capability axes (Autonomy, Generality, Planning, Memory/Persistence, Tool Economy, Self-Revision, Sociality/Coordination, Embodiment, World-Model Fidelity, Economic Throughput) aggregated by a composite AAI-Index (a weighted geometric mean). We introduce a measurable Self-Improvement Coefficient $κ$ (capability growth per unit of agent-initiated resources) and two closure properties (maintenance and expansion) that convert ``self-improving AI'' into falsifiable criteria. We specify OWA-Bench, an open-world agency benchmark suite that evaluates long-horizon, tool-using, persistent agents. We define level gates for AAI-0\ldots AAI-4 using thresholds on the axes, $κ$, and closure proofs. Synthetic experiments illustrate how present-day systems map onto the scale and how the delegability frontier (quality vs.\ autonomy) advances with self-improvement. We also prove a theorem that AAI-3 agent becomes AAI-5 over time with sufficient conditions, formalizing "baby AGI" becomes Superintelligence intuition.

With the rapid rise of large language models (LLMs) in medicine, a key question is whether they can function as competent pediatricians in real-world clinical settings. We developed PEDIASBench, a systematic evaluation framework centered on a knowledge-system framework and tailored to realistic clinical environments. PEDIASBench assesses LLMs across three dimensions: application of basic knowledge, dynamic diagnosis and treatment capability, and pediatric medical safety and medical ethics. We evaluated 12 representative models released over the past two years, including GPT-4o, Qwen3-235B-A22B, and DeepSeek-V3, covering 19 pediatric subspecialties and 211 prototypical diseases. State-of-the-art models performed well on foundational knowledge, with Qwen3-235B-A22B achieving over 90% accuracy on licensing-level questions, but performance declined ~15% as task complexity increased, revealing limitations in complex reasoning. Multiple-choice assessments highlighted weaknesses in integrative reasoning and knowledge recall. In dynamic diagnosis and treatment scenarios, DeepSeek-R1 scored highest in case reasoning (mean 0.58), yet most models struggled to adapt to real-time patient changes. On pediatric medical ethics and safety tasks, Qwen2.5-72B performed best (accuracy 92.05%), though humanistic sensitivity remained limited. These findings indicate that pediatric LLMs are constrained by limited dynamic decision-making and underdeveloped humanistic care. Future development should focus on multimodal integration and a clinical feedback-model iteration loop to enhance safety, interpretability, and human-AI collaboration. While current LLMs cannot independently perform pediatric care, they hold promise for decision support, medical education, and patient communication, laying the groundwork for a safe, trustworthy, and collaborative intelligent pediatric healthcare system.

Large Language Models (LLMs) face significant challenges in distributed healthcare, including consolidating specialized domain knowledge across institutions while maintaining privacy, reducing computational overhead, and preventing catastrophic forgetting during model updates.This paper presents a systematic evaluation of six parameter-space merging techniques applied to two architecturally compatible medical LLMs derived from the Mistral-7B base model. We introduce a novel hierarchical method that combines selective Optimal Transport (OT) alignment for attention layers with cosine similarity-weighted interpolation, designed to address permutation variance while minimizing computational overhead for edge deployment scenarios. Our study evaluates Task Arithmetic, Linear Averaging, DARE-TIES, DELLA, Breadcrumbs, and our Hierarchical approach across five medical benchmarks. Results demonstrate that architecturally compatible models benefit significantly from simple averaging methods, with Task Arithmetic achieving 45.80% accuracy on MedQA, outperforming complex pruning-based approaches. These findings offer critical insights for the deployment of distributed medical AI in resource-constrained IoT environments, where computational efficiency and model compatibility are paramount. Our work establishes that for architecturally compatible models, simple averaging provides a robust and computationally efficient baseline for knowledge consolidation, offering a pragmatic path forward for scalable medical AI systems.

How do large language models solve spatial navigation tasks? We investigate this by training GPT-2 models on three spatial learning paradigms in grid environments: passive exploration (Foraging Model- predicting steps in random walks), goal-directed planning (generating optimal shortest paths) on structured Hamiltonian paths (SP-Hamiltonian), and a hybrid model fine-tuned with exploratory data (SP-Random Walk). Using behavioural, representational and mechanistic analyses, we uncover two fundamentally different learned algorithms. The Foraging model develops a robust, map-like representation of space, akin to a 'cognitive map'. Causal interventions reveal that it learns to consolidate spatial information into a self-sufficient coordinate system, evidenced by a sharp phase transition where its reliance on historical direction tokens vanishes by the middle layers of the network. The model also adopts an adaptive, hierarchical reasoning system, switching between a low-level heuristic for short contexts and map-based inference for longer ones. In contrast, the goal-directed models learn a path-dependent algorithm, remaining reliant on explicit directional inputs throughout all layers. The hybrid model, despite demonstrating improved generalisation over its parent, retains the same path-dependent strategy. These findings suggest that the nature of spatial intelligence in transformers may lie on a spectrum, ranging from generalisable world models shaped by exploratory data to heuristics optimised for goal-directed tasks. We provide a mechanistic account of this generalisation-optimisation trade-off and highlight how the choice of training regime influences the strategies that emerge.

Medical coding converts free-text clinical notes into standardized diagnostic and procedural codes, which are essential for billing, hospital operations, and medical research. Unlike ordinary text classification, it requires multi-step reasoning: extracting diagnostic concepts, applying guideline constraints, mapping to hierarchical codebooks, and ensuring cross-document consistency. Recent advances leverage agentic LLMs, but most rely on rigid, manually crafted workflows that fail to capture the nuance and variability of real-world documentation, leaving open the question of how to systematically learn effective workflows. We present MedDCR, a closed-loop framework that treats workflow design as a learning problem. A Designer proposes workflows, a Coder executes them, and a Reflector evaluates predictions and provides constructive feedback, while a memory archive preserves prior designs for reuse and iterative refinement. On benchmark datasets, MedDCR outperforms state-of-the-art baselines and produces interpretable, adaptable workflows that better reflect real coding practice, improving both the reliability and trustworthiness of automated systems.