Deep Learning

Decomposable Transformer Point Processes (DTPP) dramatically accelerates marked point process inference by using a mixture of log-normals for inter-event times and Transformers for marks, outperformin…

DDN: Dual-domain Dynamic Normalization dynamically improves time series forecasting accuracy by addressing data distribution changes in both time and frequency domains via a plug-in module.

Data-Free Backdoor Attacks (DFBA) injects undetectable backdoors into pre-trained classifiers without retraining or architectural changes, bypassing existing defenses.

DASH combats neural network training’s plasticity loss during warm-starting by selectively forgetting memorized noise while preserving features, improving accuracy and efficiency.

CoLA: Collaborative Lifelong Adaptation boosts test-time adaptation efficiency by sharing domain knowledge across multiple devices, achieving significant accuracy gains with minimal computational over…

CRONOS: Scaling convex neural network training to ImageNet!

Credal Deep Ensembles (CreDEs) improve uncertainty quantification in deep learning by predicting probability intervals, enhancing accuracy and calibration, particularly for out-of-distribution data.

Biologically inspired Counter-Current Learning (CCL) uses dual networks for deep learning, offering comparable performance to other biologically plausible algorithms while enhancing biological realism…

This paper accelerates second-order optimization in CNNs by 4.5x, using a novel tensor network representation that simplifies convolutions and reduces memory overhead.

ControlSynth Neural ODEs (CSODEs) guarantee convergence in complex dynamical systems via tractable linear inequalities, improving neural ODE modeling.

Constrained diffusion models, trained via a novel dual approach, achieve optimal trade-offs between data fidelity and user-defined distribution constraints, enabling fairer and more controlled data ge…

Constant Acceleration Flow (CAF) drastically accelerates image generation in diffusion models by leveraging a constant acceleration equation, outperforming state-of-the-art methods in both speed and q…

CMPE: a new conditional sampler for SBI, achieves fast few-shot inference with an unconstrained architecture, outperforming current state-of-the-art algorithms on various benchmarks.

Data connectivity profoundly shapes implicit regularization in matrix factorization for matrix completion, transitioning from low nuclear norm to low rank solutions as data shifts from disconnected to…

Conformalized Time Series with Semantic Features (CT-SSF) significantly improves time-series forecasting by dynamically weighting latent semantic features, achieving greater prediction efficiency whil…

This paper introduces Selective Conformal P-Value (SCPV), a novel method for controlling FDR in conformalized multiple testing after data-dependent selection, offering a unified theoretical framework …

Conformal prediction empowers robust credal set predictions, handling aleatoric and epistemic uncertainties in classification, guaranteed to be valid with high probability!

RC3P, a novel algorithm, significantly reduces prediction set sizes in class-conditional conformal prediction while guaranteeing class-wise coverage, even on imbalanced datasets.

Boost multi-class classifier calibration by cleverly transforming the problem into a single binary calibration task!

CondTSF: One-line plugin for time series forecasting dataset condensation, boosting performance at low condensation ratios.