6745 papers
This paper proposes a principled account of out-of-distribution generalization based on feature sparsity and distribution overlap, formalizing these intuitions into a theorem that extends classic sample complexity bounds and generalizes sparse classifiers to subspace juntas.
This paper introduces MARIGOLD, a unified bi-level optimization framework that leverages zeroth-order methods to efficiently solve multi-task learning problems by dynamically balancing task gradients without requiring access to all task gradients, thereby overcoming the computational inefficiency of existing MGDA-type approaches.
This paper proposes a deterministic, seed-stable fuzzy triage system using a fine-tuned RoBERTa dual encoder to classify legal compliance and retrieve evidence with high accuracy and transparent error constraints, offering a reproducible middle ground between opaque large language models and rigid hand-crafted rules.
This paper introduces Decoupled Expected Quadratic Loss (DEQL) to generalize the EDLAE model, deriving efficient closed-form solutions for the previously unexplored hyperparameter range that empirically outperform the original baseline on benchmark datasets.
This paper introduces the information-theoretic concept of Context Channel Capacity () to explain catastrophic forgetting in continual learning, proving that zero forgetting requires and demonstrating that architectures with structural context pathways (like HyperNetworks) bypass the Impossibility Triangle to achieve near-perfect retention, whereas methods lacking such capacity inevitably suffer significant forgetting.
DualSpec is a heterogeneous speculation framework that accelerates deep research agents by distinguishing between high-uncertainty Search and low-uncertainty Visit actions to enable a lightweight, confidence-based verification process, achieving up to 3.28 speedup without compromising accuracy.
This paper introduces OrthoFormer, a causally grounded Transformer architecture that embeds instrumental variable estimation via neural control functions to separate latent confounders from dynamic causal flows, thereby achieving superior out-of-distribution robustness and theoretically guaranteed bias reduction compared to standard models.
This paper addresses the underexplored challenge of generalization in online reinforcement learning for mobile GUI agents by introducing the AndroidWorld-Generalization benchmark and a scalable GRPO-based training system, demonstrating that while RL significantly improves zero-shot performance on unseen task instances, generalization to new templates and applications remains difficult and benefits from test-time few-shot adaptation.
The paper proposes Data Agent, an end-to-end dynamic data selection framework that formulates sample selection as a training-aware sequential decision-making problem to accelerate model training while preserving performance across diverse datasets and tasks.
This paper establishes finite-sample guarantees for cost-driven state representation learning in infinite-horizon time-invariant Linear Quadratic Gaussian (LQG) control by analyzing two approaches—explicit latent modeling and implicit MuZero-like dynamics—while introducing a key technical proof of persistency of excitation for a novel stochastic process arising from quadratic regression.
The paper proposes PACT, a fine-tuning framework that preserves LLM safety alignment by selectively constraining the model's confidence on a small subset of safety-related tokens during training, thereby preventing alignment drift without compromising downstream task performance.
This paper demonstrates that a deliberately simplistic discrete tokenization strategy, combined with adaptive Gaussian smoothing, enables Transformers to outperform tuned gradient boosting models on large-scale tabular forecasting tasks while producing well-calibrated probability distributions.
This paper introduces Dial, a knowledge-grounded framework that addresses the challenges of generating executable SQL across heterogeneous database systems by employing dialect-aware logical planning, a hierarchical intent-aware knowledge base, and an execution-driven debugging loop, achieving significant improvements in translation accuracy and dialect feature coverage on the newly constructed DS-NL2SQL benchmark.
The paper introduces SLNet, a super-lightweight 3D point cloud recognition network utilizing Nonparametric Adaptive Point Embedding (NAPE) and Geometric Modulation Units (GMU) to achieve state-of-the-art accuracy on benchmarks like ModelNet40 and ScanObjectNN with significantly fewer parameters and computational costs compared to existing models.
This paper introduces the Dual-Stream Transformer, an architecture that decomposes the residual stream into separate token and context streams with tunable mixing strategies to achieve a balance between high interpretability and minimal performance loss while demonstrating robustness to attention amplification.
The paper introduces AFTUNE, a framework that ensures the integrity of cloud-based large language model fine-tuning and inference by employing a lightweight recording and spot-check mechanism to generate verifiable execution traces, thereby enabling clients to practically audit proprietary AI processes without prohibitive overhead.
This monograph offers a rigorous theoretical and methodological overview of probabilistic inference and learning with Stein's method, detailing the construction and properties of Stein discrepancies, their connection to Stein variational gradient descent, and providing precise definitions and proofs.
This paper proposes a lightweight, self-supervised framework that augments a frozen speech enhancement backbone with low-rank adapters, enabling efficient on-device adaptation to dynamic real-world noise conditions by updating fewer than 1% of parameters while achieving significant signal quality improvements.
This paper introduces a conditional diffusion-transformer framework that generates diverse ensembles of 3D *E. coli* genome conformations guided by Hi-C contact maps, effectively reconstructing heterogeneous structures whose ensemble averages align with experimental data while preserving conformational diversity.
This paper proposes and validates the Late Fusion Architecture (LFA), a transformer variant that enforces interpretability by design through architectural stream independence, which separates symbolic and contextual processing to prevent premature entanglement and significantly improve model stability and functional modularity compared to standard transformers.