Huey Sun

Huey Sun, Anabel Yong, Lorenzo Gilly et al.

Contrastive decoding is a lightweight and effective inference-time method that improves the quality of text generation in Large Language Models. However, algorithms such as DoLa (Decoding by Contrastive Layers) have only been implemented in decoder-only architectures and studied for their impact on improving factuality. This work adapts DoLa for the T5 and FLAN-T5 model families and evaluates its impact on the models' instruction following capabilities, which to our knowledge is the first implementation of a contrastive decoding strategy in an encoder-decoder architecture. Our results show that DoLa improves the faithfulness of text generation for certain categories of tasks and harms others. To understand these results, we present a layer-by-layer analysis of logit evolution in a FLAN-T5 model to quantify DoLa's impact on token output probabilities.

Oli Bridge, Huey Sun, Botond Branyicskai-Nagy et al.

While transformer-based architectures have taken computer vision and NLP by storm, they often require a vast amount of parameters and training data to attain strong performance. In this work, we experiment with three distinct pre-training, intermediate fine-tuning, and downstream datasets and training objectives to explore their marginal benefits on a small 5M-parameter vision transformer. We find that while pre-training and fine-tuning always help our model but have diminishing returns, intermediate fine-tuning can actually show harmful impact on downstream performance, potentially due to dissimilarity in task mechanics. Taken together, our results suggest that small-scale ViTs benefit most from targeted pre-training and careful data selection, while indiscriminate stacking of intermediate tasks can waste compute and even degrade performance.

Jake Williams, Abel Tadesse, Tyler Sam et al.

Transfer learning involves taking information and insight from one problem domain and applying it to a new problem domain. Although widely used in practice, theory for transfer learning remains less well-developed. To address this, we prove several novel results related to transfer learning, showing the need to carefully select which sets of information to transfer and the need for dependence between transferred information and target problems. Furthermore, we prove how the degree of probabilistic change in an algorithm using transfer learning places an upper bound on the amount of improvement possible. These results build on the algorithmic search framework for machine learning, allowing the results to apply to a wide range of learning problems using transfer.

Tyler Sam, Jake Williams, Abel Tadesse et al.

Previous studies have used a specific success metric within an algorithmic search framework to prove machine learning impossibility results. However, this specific success metric prevents us from applying these results on other forms of machine learning, e.g. transfer learning. We define decomposable metrics as a category of success metrics for search problems which can be expressed as a linear operation on a probability distribution to solve this issue. Using an arbitrary decomposable metric to measure the success of a search, we demonstrate theorems which bound success in various ways, generalizing several existing results in the literature.