Few-Shot Learning

LLM-powered Tri-level learning framework enhances time series OOD generalization.

Transformers surprisingly learn second-order optimization methods for in-context linear regression, achieving exponentially faster convergence than gradient descent!

Tiny Time Mixers (TTMs) achieve state-of-the-art zero/few-shot multivariate time series forecasting, outperforming existing benchmarks while drastically reducing computational requirements.

Test-time adaptation strengthens the linear correlation between in- and out-of-distribution accuracy, enabling precise OOD performance prediction and hyperparameter optimization without labeled OOD da…

On-device training with fixed-point forward gradients enables efficient model personalization on resource-constrained edge devices, overcoming backpropagation’s memory limitations.

Prospective Representation Learning (PRL) revolutionizes non-exemplar class-incremental learning by proactively reserving embedding space for new classes and minimizing the shock of new data on previo…

CoDA-NO, a novel neural operator, revolutionizes multiphysics PDE solving via codomain tokenization, enabling efficient self-supervised pretraining and few-shot learning for superior generalization.

Pin-Tuning: A parameter-efficient method for few-shot molecular property prediction that significantly improves accuracy with fewer trainable parameters via in-context tuning and Bayesian weight cons…

OTTER effortlessly adapts zero-shot models to new tasks by adjusting predictions using optimal transport, improving accuracy significantly without extra training data.

Boosting Robust Few-Shot Learning with Adversarial Meta-Tuning!

CoPA improves cross-domain few-shot learning by adapting separate transformations for prototype and image embeddings, significantly enhancing performance and revealing better representation clusters.

Meta-Exploiting Frequency Prior enhances cross-domain few-shot learning by leveraging image frequency decomposition and consistency priors to improve model generalization and efficiency.

aTLAS: a novel parameter-efficient fine-tuning method using learned anisotropic scaling of task vectors for enhanced knowledge composition and transfer.

Multi-head transformers utilize distinct attention patterns across layers—multiple heads are essential for initial data preprocessing, while a single head suffices for subsequent optimization steps, o…

Researchers developed FSAFW, a novel framework generating universal adversarial perturbations effective against various Few-Shot Learning paradigms, surpassing baseline methods by over 16%.

Few-shot task learning through inverse generative modeling (FTL-IGM) enables AI agents to quickly master new tasks from minimal data by leveraging invertible generative models.

Few-shot diffusion models efficiently generate customized images; this paper provides the first theoretical explanation, proving improved approximation and optimization bounds, escaping the curse of d…

Fast Graph Sharpness-Aware Minimization (FGSAM) accelerates few-shot node classification by cleverly combining GNNs and MLPs for efficient, high-performing training.

EPIC: Effective prompting makes LLMs generate high-quality synthetic tabular data, significantly boosting imbalanced-class classification.

D2R2: A novel diffusion-based model for tabular few-shot learning, achieves state-of-the-art results by leveraging semantic knowledge and distance matching.