All papers

This paper provides minimax rates and limits for recalibrating predictions across many binary questions when each question has few ground-truth labels.

This paper introduces a risk-seeking optimization method to prepare target quantum states with high fidelity on noisy bosonic quantum hardware.

This paper proposes LlMBO, applying Bayesian optimization to ultra-low-dimensional fine-tuning as an alternative to gradient-based policy optimization.

This paper analyzes heterogeneous asynchronous multi-player bandits, proves the centralized allocation problem is NP-hard, and studies greedy approximation policies.

This paper introduces CACTUS, an ADMM-based optimizer that speeds up curvature-aware pruning and preserves influence functions in compressed language models.

This paper replaces squared Euclidean distance with Euclidean distance in Flow Matching to improve distributional robustness against data contamination.

The paper derives explicit variance approximations to extend circulant permutation techniques to different MinHash estimators.

The paper provides last-iterate convergence bounds for clipped stochastic gradient descent under heavy-tailed noise and causal adversarial perturbations.

The paper establishes linear last-iterate convergence rates for optimistic learning algorithms in multiplayer harmonic games.

The paper proposes a flow-matching model that integrates deterministic forecasting as a velocity component for stable probabilistic time-series predictions.

The paper derives minimax risk bounds showing how covariate design smoothness affects pointwise conditional average treatment effect estimation.

The paper presents a profiling-guided framework to allocate precision budgets adaptively across layers and tokens during KV cache quantization.

The paper introduces a training-free conditional sampling method for pretrained Schrödinger bridge generative models.

The paper demonstrates that the placement timing of curvature updates in Kronecker-preconditioned optimizers significantly affects language model training loss.

The paper proves theoretical teaching complexity bounds for closed neighborhoods in planar graphs without collusion.

The paper compares transformer sequence predictions against an optimal Bayesian predictor across structured Markov sources.

The paper proposes a reinforcement learning method to selectively transfer data for carbon-footprint modeling in new manufacturing settings.

The paper presents an adaptive, training-free speculative decoding method to speed up inference for vision-language models.

The paper develops an off-policy reinforcement learning method with token-wise importance sampling and proximal KL regularization.

The paper analyzes how network depth influences the internal loop geometry and spatial reach of gauge-equivariant graph neural networks.

The paper formalizes layer-to-layer representation sharing as a memory storage and retrieval framework for deep transformer networks.

The paper analytically characterizes the mechanisms and conditions behind the grokking phenomenon in a simplified learning setup.

The paper presents a neural network architecture for simultaneous model selection and parameter posterior estimation in chemical reaction networks.

The paper introduces methods to generate graphs whose topological persistence diagrams match target descriptors exactly rather than approximately.

The paper presents a method for training diffusion models in lattice field theory directly from their own generated samples.

The paper introduces a framework that uses language models to extract task-scheduling priority functions from optimal solver solutions on directed acyclic graphs.

The paper distills an LLM-based search relevance judge into smaller multimodal student models for efficient e-commerce search evaluation.

The paper proposes a continuous probabilistic Gaussian mixture framework to improve human pose estimation under dense occlusions without retraining models.

This paper evaluates category name matching versus pairwise geometry metrics for candidate retrieval in context-constrained 3D vision-language models.

This paper evaluates calibration differences between object-level softmax and agreement metrics when resolving ambiguous 3D referring queries.

This paper presents a flow-matching transformer pipeline to generate full gigapixel histopathology whole-slide images.

This paper studies transformer architectural choices for handling missing modalities and longitudinal observations in multimodal healthcare datasets.

This paper presents a unified model for dense visual correspondence across 25 pairs of multi-modal sensor inputs.

The paper presents a four-gate verification protocol to evaluate whether audio-conditioned visual prediction models genuinely rely on input audio signals.

The paper introduces a lightweight test-time attention framework that leverages single-photon sensor noise as an informative prior for indoor depth estimation.

The paper presents a theoretical framework using nesting properties to calculate distance bounds and compare language model truncation sampling algorithms efficiently.

This paper introduces prototype-purity pseudo-labeling to extract accurate instance-level annotations from weakly supervised whole-slide pathology images.

This paper introduces a training-free caching method that separates cache synchronization and computation repair to accelerate autoregressive video diffusion generation.

Proposes a knowledge distillation method that focuses learning signals on specific structural disagreements between teacher and student representations.

Evaluates router-learning strategies for combining domain-specialized diffusion experts in multi-domain medical image generation.

Proposes a training-free frame packing method enabling vision-language models to process long videos with constant memory consumption.

The paper introduces an importance-weighted flow matching approach to explicitly learn powered conditional distributions without standard classifier-free guidance.

The paper introduces a multimodal entity linking method that combines vision-language models with heterogeneous graph networks for cultural heritage data.

The paper proposes a closed-form variational filtering algorithm for the last layer of deep classifiers to improve uncertainty calibration.

The paper introduces a model and dataset for financial candlestick reasoning by generating and executing code to gather visual evidence.

The paper presents a diffusion transformer operating on per-frame velocities to generate rigid point trajectories from a single image.

The paper proposes a structured pruning method to compress vision foundation models into compact task-specific models under parameter budgets.

The paper proposes a training-free guidance method enabling continuous valence-arousal emotional control in foundation image editors.

The authors introduce a hybrid architecture combining EfficientNetV2 and Swin Transformer to classify Alzheimer's and Parkinson's disease from brain MRI.

Proposes a post-training decoder alignment method for latent generative models to resolve distribution mismatches between training and generation latents.

Investigates active learning dynamics by regularizing representation collapse and managing sample acquisition across training iterations.

Proposes a calibrated disambiguation model to optimize sample weighting when training classifiers on datasets with noisy or excessive labels.

Introduces a deficit-driven stratified acquisition algorithm to create a balanced, co-registered satellite imagery corpus for self-supervised Earth observation.

Proposes a differentiable graph-level loss function to measure semantic and structural discrepancies between predicted 3D scene graphs and ground truth.

Introduces a low-rank adaptation method that splits rank budgets into stable and dynamic branches to capture residual gradients outside the tangent space.

This paper proposes a hierarchical framework for learning whole-arm robot contact manipulation tasks from human demonstrations.

This paper introduces an active learning framework balancing uncertainty with submodular spatial coverage for chest X-ray lesion segmentation.

This paper presents a lightweight, interpretable neural network architecture for resource-constrained time-series classification on microcontrollers.

This paper proposes a multimodal variational autoencoder framework that separates task-relevant information to improve clinical predictions even with missing modalities.

Examines industrial logical anomaly detection under capture shifts and proposes a reference-free classifier trained on synthetic anomalies.