This paper studies generalization error bounds for Transformer models using offset Rademacher complexity. The core idea is to derive sharp excess risk bounds that achieve optimal $O(1/n)$ convergence rates---improving upon the standard $O(1/\sqrt{n})$---for single-head, multi-head, and multi-layer architectures, with explicit dependence on matrix ranks and parameter norms. The authors further extend these results to unbounded sub-Gaussian and heavy-tailed input distributions, broadening the applicability beyond standard boundedness assumptions.

High-fidelity CFD for vehicle aerodynamic drag is bottlenecked not by solver wall time but by workflow friction—CAD cleanup, meshing retries, and queue contention. This paper proposes a contract-centric blueprint where self-evolving coding agents search over executable surrogate programs (not static models) to predict drag coefficient $C_d$ under industrial constraints. The system combines Famou-Agent-style evaluator feedback with population-based island evolution and hard evaluation contracts that enforce leakage prevention, deterministic replay, and resource budgets, aiming for a screen-and-escalate deployment where uncertain cases trigger automatic fallback to high-fidelity CFD.

Rule-State Inference (RSI) addresses compliance monitoring in domains like taxation where authoritative rules are known a priori but observations are partial, noisy, or strategically distorted. The paper proposes a Bayesian framework that inverts the standard ML paradigm: instead of learning rules from data, RSI encodes regulatory rules as structured priors and infers latent rule states (activation, compliance rate, parametric drift) via posterior inference. This enables zero-shot compliance assessment without labeled training data—a critical capability for low-resource environments where non-compliance labels are scarce or legally sensitive.

MIST addresses the challenge of generating high-quality SQL test cases for Database Management Systems using lightweight Large Language Models. The framework combines a feature-guided synthesis stage that leverages hierarchical documentation structures with error feedback, and a Monte Carlo Tree Search-based mutation stage to overcome coverage plateaus. This two-pronged approach aims to achieve high code coverage in resource-constrained industrial environments where only small LLMs can be deployed locally.

This paper addresses robotic reaching in cluttered, unseen environments using a hybrid rigid-soft continuum manipulator. The core idea is a real-time pipeline that combines multi-view RGB reconstruction (Mast3r), open-world object detection (YOLO-World), shape-aware RRT* planning with asymmetric collision constraints, and a learned controller trained on pose-to-actuation data. If validated at scale, this could enable robots to navigate dense foliage or disaster debris where rigid arms fail and pure soft arms lack reach.

The paper critiques the institutionalization of LLM benchmarks as "Silicon Bureaucracy" and "AI Test-Oriented Education", arguing high scores often conflate exam-oriented competence with genuine generalization due to data contamination. It proposes an audit framework using a router-worker setup: clean-control routers transmit full questions while noisy routers delete, rewrite, and perturb before aggregation. For clean benchmarks, noisy aggregation should not systematically exceed the baseline; persistent above-baseline gains suggest contamination-related memory activation. The core finding—that 10 of 12 models exceed clean baselines under multi-router noisy conditions—challenges the interpretability of raw benchmark scores.

DiT-Flow tackles multi-condition speech enhancement (noise, reverberation, codec compression) by combining flow matching with a latent Diffusion Transformer (DiT) backbone. The paper proposes operating flow matching in a VAE-compressed latent space for efficiency, introduces StillSonicSet (a synthetic dataset with realistic room acoustics for stationary sources), and applies Mixture-of-LoRA-Experts (MoELoRA) for parameter-efficient adaptation to unseen distortions. The work matters because most SE models fail when deployed on real-world audio with compound distortions unseen during training.

CAID tackles long-horizon software engineering tasks where single agents struggle with accuracy and wall-clock time. The core idea is Centralized Asynchronous Isolated Delegation: a manager decomposes tasks into dependency graphs and delegates to multiple engineer agents working in isolated git worktrees, integrating progress via branch-and-merge. The system improves accuracy by 26.7% absolute on PaperBench and 14.3% on Commit0, demonstrating that structured coordination grounded in SWE primitives outperforms simply scaling single-agent iteration budgets.

AdaRubric solves the static-rubric bottleneck in LLM-as-Judge evaluation by dynamically generating task-specific evaluation dimensions from task descriptions. It scores agent trajectories step-by-step with confidence-weighted per-dimension feedback and filters preference pairs using the DimensionAwareFilter—a provably necessary mechanism to prevent high-scoring dimensions from masking failures. The approach achieves Pearson $r=0.79$ correlation with human judgments and yields substantial downstream gains: +6.8–8.5 percentage points in DPO task success and +6.6 pp faster PPO convergence at 5K steps.

The paper tackles instability in multi-agent reinforcement learning for LLM reasoning, where noisy, heavy-tailed rewards break standard GRPO batch-mean normalization. It proposes DACR, a structured Answer-Critique-Rewrite protocol with cross-improvement rewards, and ARE, a robust estimator that replaces empirical means with a Median-of-Means variant using adaptive losses. Experiments on mathematical reasoning and aerial vision-language navigation demonstrate improved accuracy and training stability under synthetic noise contamination.

RAG improves factual reliability but doesn't eliminate hallucinations. The paper reveals a mechanistic paradox: induction heads that copy correct answers from context simultaneously trigger entropy neurons that suppress confidence, causing entropy-based uncertainty signals to fail. INTRYGUE gates predictive entropy using induction head activation (SinkRate) to correct this inflation, offering a training-free method for reliable RAG hallucination detection.

This paper applies the classic Abernathy-Utterback (A-U) innovation diffusion model to generative AI's environmental impact. The authors argue that alarmist predictions about GAI's carbon footprint often ignore how innovation diffusion drives process optimization and efficiency gains. They forecast that the GAI industry is transitioning from the 'fluid' A-U:1 phase to the 'transitional' A-U:2 phase, where dominant designs will emerge. The paper predicts two main business models: large generalist platforms serving mass audiences, and smaller specialized models targeting specific use cases. Their core argument is that GAI 'will never be green, but its impact may not be as problematic as is sometimes claimed' depending on which business model dominates.

Concrete mix design requires balancing competing objectives of mechanical strength and sustainability. BOxCrete introduces a Gaussian Process regression framework trained on 533 strength measurements from 123 unique mixtures to predict compressive strength evolution over curing time and optimize mixes for embodied carbon using multi-objective Bayesian Optimization. The work addresses a critical gap in the literature by providing an open-source alternative to proprietary industrial datasets and models.

This paper addresses computational barriers for Brazilian Portuguese question answering by systematically evaluating Parameter-Efficient Fine-Tuning (PEFT) methods on BERTimbau models using the SQuAD-BR dataset. The authors test LoRA, DoRA, QLoRA, and QDoRA across Base (110M) and Large (335M) variants, demonstrating that LoRA achieves 95.8% of full fine-tuning performance while reducing training time by 73.5%. A key finding is that PEFT methods require substantially higher learning rates ($2\times 10^{-4}$) than standard BERT fine-tuning to achieve optimal results, with quantization resilience favoring larger models.

This paper tackles the problem of measuring dialectal bias in LLMs for Bengali, a low-resource language with nine major regional variants. The authors propose a two-phase framework combining RAG-based translation to create dialectal benchmarks with an RLAIF-inspired evaluation protocol that uses CoT-first reasoning and multi-judge validation. They expose the catastrophic failure of traditional metrics like BLEU and WER for agglutinative dialectal Bengali, showing that LLM-as-judge better predicts human quality assessments.

CataractSAM-2 adapts Meta's Segment Anything Model 2 (SAM-2) for real-time semantic segmentation in cataract surgery videos. The core idea is to fine-tune only the prompt encoder and mask decoder while freezing the image encoder, enabling precise segmentation of anatomical structures and surgical instruments under challenging conditions like glare and occlusion. The paper also introduces an interactive annotation framework that propagates sparse user prompts across video frames to accelerate ground-truth generation.

Large Language Models often inherit societal biases that manifest as stereotyped associations across demographic groups. This paper proposes CatRAG, a dual-mechanism debiasing framework that combines a category-theoretic functor-guided projection—collapsing protected-attribute directions in embedding space via spectral decomposition—with diversity-aware Retrieval-Augmented Generation to ground inference in balanced evidence. Evaluated on the BBQ benchmark across Llama-3, GPT-OSS, and Gemma-3, the method claims to reduce bias scores from ~60% to near zero while improving accuracy by up to 40% over base models.

The paper tackles the inefficiency of homogeneous compute allocation in multi-task supervised fine-tuning (SFT), where fast-learning tasks overfit while slow ones remain under-trained. The authors propose mSFT, an iterative algorithm that dynamically excludes overfitting sub-datasets and reverts to optimal checkpoints. Their approach consistently outperforms baselines across 6 models and 10 benchmarks, sometimes reducing compute while improving accuracy.

Unified-MAS tackles a critical failure mode in automatic Multi-Agent Systems: their severe performance degradation in knowledge-intensive domains like healthcare and law, where general-purpose reasoning nodes fall short. The core innovation decouples granular node implementation from topological orchestration through an offline two-stage pipeline that synthesizes domain-specific agent nodes via external knowledge retrieval and refines them using a perplexity-guided reward signal. This paradigm matters because it promises to catapult general-purpose Auto-MAS to expert-level performance without costly manual engineering of domain-specific agents.

AgentHER tackles the data waste problem in LLM agent training by adapting Hindsight Experience Replay (HER) from RL to natural-language trajectories. The core insight is that failed trajectories—typically 60–75% of collected data—often represent valid demonstrations for achievable alternative goals. The paper proposes a four-stage pipeline with multi-judge verification that converts discarded failures into SFT and DPO training data, yielding +7.1–11.7 pp gains over success-only fine-tuning across four model families on WebArena and ToolBench.