All papers

Replaces discrete attention lookup in transformers with integration over a continuous memory curve to improve validation likelihood.

Analyzes the price of anarchy and economic efficiency of first-price auctions where AI agents bid to serve user prompts.

Presents a retraining-free method to adapt transformer sizes dynamically across diverse computational budgets using joint soft pruning of components.

Introduces multi-step action forecasting targets alongside actor rollouts to train language agents more effectively from intermediate trajectory steps.

Proposes a hierarchical residual policy learning framework to fine-tune mobile robot manipulation policies with decoupled base and arm adjustments.

Generatively models humanoid robot control policy weights using a latent diffusion framework combined with a linear autoencoder and Diffusion Transformer.

This paper evaluates the ability of language models to generate non-functional requirements from functional software specifications.

This paper combines evolutionary computation with language models to optimize register-transfer level hardware code for correctness and efficiency.

This paper introduces specification templates to define operational boundaries, authority, and oversight for autonomous software agents.

The paper introduces an architecture that separates safe and unsafe visual knowledge during pretraining so unsafe branches can be disabled.

The paper proposes a physics-informed forward-inverse machine learning framework for sparse recipe formulation in industrial dyeing.

The paper develops a graph attention network to transfer prototype facade articulation patterns onto building massing targets.

The paper develops an early-warning pipeline using public balance-sheet filings to predict defined-benefit pension sponsor distress.

The paper introduces a hierarchical multi-agent reasoning system combining retrieval and knowledge graphs to find chemical material substitutes.

The paper presents a theoretical negative result on physically implementing reinforcement-learning-driven reasoning in certain physical neural networks.

The paper provides a stochastic analysis perspective linking score matching, Langevin sampling, and diffusion model time reversal.

The paper introduces a lossless compression codec for model checkpoint trajectories conditioned on reference floating-point exponents.

The paper proposes a dynamic stratified grid sampling approach to train implicit neural representations on large scientific data faster.

The paper proposes an embedding-based graph kernel entropy method for computationally efficient hallucination detection in language models.

The paper extends recursive linear structural equation model identification to non-Gaussian noise using conditional prediction risk.

The paper proposes training looped transformers at variable depths only during a short finishing stage to reduce compute costs.

The paper examines importance estimate shrinkage to mitigate catastrophic forgetting in regularization-based continual learning.

The paper establishes theoretical conditions for identifying individual content factors in self-supervised learning via designed augmentation strengths.

The paper introduces a client-side reinforcement learning scheduler to optimize synchronization decisions in edge federated learning.

The paper probes convolutional neural networks to measure how much layer input information can be reconstructed from layer outputs.

The paper proposes an active robot visual learning framework driven by self-supervised curiosity and world model prediction errors.

The paper proves convergence guarantees for decentralized bandit learning in infinite-horizon, general-sum Markov games.

The paper presents a trust-region Bayesian optimization framework designed for noisy, high-dimensional blackbox problems.

The paper presents a continuous-time neural ODE solver for simulating cloth dynamics at arbitrary timestamps.

The paper analyzes depth transfer of hyperparameters in neural networks using an L-smoothness theoretical framework.

The paper presents a conformal prediction method to reliably infer Signal Temporal Logic formulas for multiclass time series classification.

The paper bounds the sample and query complexity of multicalibration using an empirical risk minimization oracle.

The paper proposes a conformal calibration layer to provide finite-sample control over false positives in language model classification workflows.

The paper tests predictive neural control against classical controllers in simulated spiking neural populations under varying sensitivity conditions.

The paper investigates dynamic tensor rematerialization, identifying cases where increasing memory budgets causes training to unexpectedly run out of memory.

The paper presents an adaptation loss incorporating return direction and price trajectory calibration for test-time financial time series forecasting.

The paper proposes a probabilistic decomposition of neural models that separates predictable features from an input-dependent reliability score.

The paper applies a sparsity penalty to task representations to sharpen reward signals in zero-shot reinforcement learning.

The paper introduces a backpropagation-free training method for generative relaxation models suitable for analog neuromorphic hardware execution.

The paper examines using transfer probes to group business tasks when sharing Low-Rank Adaptation experts under fixed parameter budgets.

The paper studies depth extrapolation in looped transformers for graph connectivity, identifying finite iterations beyond which test performance degrades.

The paper presents a mass-conserving neural network architecture designed for interpretable rainfall, evapotranspiration, and runoff simulation across diverse river catchments.

The paper introduces an adapter method that disentangles subject identity from task content during downstream fine-tuning of brainwave foundation models.

The paper develops identification theory and asymptotic estimation methods for nonsymmetric determinantal point processes capable of modeling positive dependence.

The paper presents a physics-constrained generative latent operator framework to simulate multiscale particle accelerator beam dynamics efficiently.

The paper proposes a method to align and merge multiple Low-Rank Adaptations using a shared basis guided by pretrained model weights.

The paper introduces a chunk-wise model-based reinforcement learning approach that uses single-pass rollouts to optimize policies without recurrent backpropagation through time.

The paper derives strong Gaussian approximations for constant learning rate stochastic gradient descent under heavy-tailed noise to enable trajectory-level inference.

The paper evaluates carbon-aware scheduling for geographically distributed language model serving under strict latency service-level objectives using an empirical simulator.

The paper combines language models with Bayesian optimization to guide architectural design exploration for heterogeneous chiplet systems.

The paper proposes a transferable spatiotemporal graph learning framework to forecast road network disruption caused by extreme rainfall across unseen regions.

The paper introduces a Bayesian optimization method that combines language model relational priors with structural descriptors for mixed-variable scientific discovery.

The paper analyzes policy gradient updates that divide trajectory scores by length, identifying mathematical conditions under which expected returns can decrease.

The paper presents an evidence admission framework that verifies anomalies before updating the persistent principles of scientific language model agents.

The paper proposes patchwise adaptive normalization to represent one-way coupled exogenous fields in transformer surrogates for partial differential equations.

This paper presents a system that speeds up multi-model language model serving by translating and reusing attention maps across models.

This paper identifies spatial boundary biases in Bayesian optimization and modifies exploration using parameterized spatial weight fields.

This paper introduces a linear-time algorithm to place codec operators in computational graphs to compress operand traffic during on-device inference.

This paper proposes a decomposed routing framework that distills large language model knowledge into lightweight graph neural networks.

This paper analyzes the complexity and theoretical benefits of meeting-triggered information exchange in multi-token random-walk decentralized learning.