Machine Learning

Computational models have become essential tools for understanding signalling networks and their non-linear dynamics. However, these models are typically constructed manually using prior knowledge and can be over-reliant on study bias. Scaling up the construction and analysisof models to take advantage of increasingly abundant ‘omics data can bridge these gaps by providing a comprehensive view of signalling events and how they influence cellular phenotypes.

Sparse AutoEncoder (SAE) latents show promise as a method for extracting interpretable features from large language models (LM), but their overall utility as the foundation of mechanistic understanding of LM remains unclear. Ideal features would be linear and independent, but we show that there exist SAE latents in GPT2-Small display non-independent behaviour, especially in small SAEs. We investigate: 1. What fraction of SAE latents might best be understood in groups rather than individually? 2. Do low-dimensional subspaces mapped by co-occurring groups of features ever provide a better unit of analysis (or possibly intervention) than individual SAE latents?

Making decisions on how best to treat cancer patients requires the integration of different data sets, including genomic profiles, …