Transfer Learning

Linear model transfer learning achieves universal generalization error improvements, depending only on first and second-order target statistics, and defying Gaussian assumptions.

This paper presents efficient algorithms for transfer learning in latent variable network models, achieving vanishing error under specific conditions, and attaining minimax optimal rates for stochasti…

This work unveils a causal lens on extrapolation, offering theoretical guarantees for accurate predictions on out-of-support data, even with limited target samples.

A new dual-learning model resolves the paradox of perceptual learning, showing how task-based and feature-based learning interact to produce both specific and transferable improvements in sensory perc…

Lowering a neural network’s geometric complexity during pre-training enhances neural collapse and improves transfer learning, especially in few-shot scenarios.

TFGDA: Leveraging graph topology and feature alignment for superior semi-supervised domain adaptation.

SAUE: A novel multi-source blended-target domain adaptation approach using style adaptation and uncertainty estimation to improve model robustness and accuracy.

Accelerate unpaired data translation with Schrödinger Bridge Flow, a novel algorithm solving optimal transport problems efficiently without repeatedly training models!

I-Div accurately quantifies distribution discrepancy between training and test datasets without test labels, enabling reliable hypothesis applicability evaluation in complex scenarios.

Reinforced Cross-Domain Knowledge Distillation (RCD-KD) dynamically selects target samples for efficient knowledge transfer from a complex teacher model to a compact student model, achieving superior …

This paper introduces a novel method for creating highly accurate and narrow prediction intervals even when data distribution shifts unexpectedly, significantly improving machine learning model reliab…

Improved unsupervised domain adaptation framework achieves superior performance via refined f-divergence and novel f-domain discrepancy, enabling faster algorithms and tighter generalization bounds.

Scalable SEEG decoding model, seegnificant, leverages transformers to decode behavior across subjects despite electrode variability, achieving high accuracy and transfer learning capability.

MCTS-transfer: Iteratively refining Bayesian optimization via Monte Carlo tree search for efficient black-box optimization using transfer learning.

TABULA-8B, a novel language model for tabular prediction, achieves state-of-the-art zero-shot and few-shot performance across various benchmarks, exceeding existing methods by 5-15 percentage points.

Multi-task learning and finetuning show surprising feature reuse biases, including a novel ’nested feature selection’ regime where finetuning prioritizes a sparse subset of pretrained features, signif…

This paper enhances learnware dock systems by using model outputs to improve heterogeneous learnware management, enabling effective task handling even without perfectly matched models.

GFT: a novel graph foundation model using transferable computation trees as tokens, improving generalization and reducing negative transfer in graph learning.

GOPSA: a novel geodesic optimization method significantly improves cross-site age prediction from EEG data by jointly handling shifts in data and predictive variables.

Fine-tuning pre-trained models often degrades performance on unseen classes. This work reveals that the problem stems from logit scale discrepancies, not feature loss, and shows that post-processing c…