The paper explores the use of prompts that build synthetic training data with specific, documented attributes using large language models (LLMs) for natural language processing (NLP) tasks. The authors present an empirical study on data generation encompassing aspects like bias, diversity, and efficiency.

The paper explores the use of pseudolabels, which are heuristic labels for unlabeled data, to enhance the performance of vision-language models like CLIP via prompt tuning. The authors investigate different learning paradigms and prompt modalities and find that iterative prompt-training strategies leveraging CLIP-based pseudolabels lead to significant improvements in CLIP’s image classification performance.

The paper introduces Alfred, a system for programmatic weak supervision (PWS) that creates training data for machine learning by prompting. It enables users to encode their subject matter expertise via natural language prompts for language and vision-language models.

The paper proposes a statistical label model called FABLE that incorporates instance features to improve the accuracy of inferred truth in Programmatic Weak Supervision (PWS). FABLE is built on a mixture of Bayesian label models, where the coefficients of the mixture components are predicted by a Gaussian Process classifier based on instance features.

This paper proposes generalizations of CWOLA and SALAD, which exploit multiple reference datasets to improve performance in resonant anomaly detection, and provides finite-sample guarantees to go beyond existing asymptotic analyses.

This paper proposes “Ask Me Anything” (AMA), a prompting method that uses weak supervision to combine noisy predictions from multiple prompts generated from an LLM, resulting in an average 10.2% performance lift over the few-shot baseline across a variety of different open-source models.

The authors propose Contrastive Adapting, an efficient adapter training strategy that improves the group robustness of large pretrained foundation models (FMs) without finetuning, leading to up to 56.0 percentage points of increase in accuracy compared to zero-shot.

Zero-shot learning with Common Sense Knowledge Graphs is a general-purpose framework with a novel transformer graph convolutional network for generating class representations from common sense knowledge graphs, which improves over existing WordNet-based methods on zero-shot learning tasks.