Colloquia and Seminars

In meta reinforcement learning (meta-RL) an agent seeks an optimal policy when facing a new unseen task that is sampled from a known task distribution. Such a policy leads an effective trade-off between information gathering and reward accumulation. The offline variant of meta-RL (OMRL) presents a challenge to learn such a policy, as previous work established an identifiability problem in OMRL termed MDP ambiguity. This problem relates to the difficulty of learning a neural network that can infer the task at hand at test time. We propose a new method to utilize prior knowledge of the task distribution to mitigate the identifiability problem in OMRL. Additionally, we propose a novel method to evaluate an inference model \textit{offline}, which is more efficient and accurate than the online alternative of policy optimization. Finally, we show that the offline version of the popular VariBAD algorithm can learn a suboptimal representation for task inference, and propose a simple modification that uses contrastive predictive coding to improve its performance. We compare our methods to Offline VariBAD on two ambiguity-prone tasks and demonstrate results that are on par or better than policy replay - a state of the art method for solving MDP ambiguity - while requiring weaker assumptions.

Over the past two decades, advancements in gene expression laboratory methods have brought about a level of maturity that allows for the routine examination of gene expression at both the single-cell level and spatially across tissues. However, existing data analysis methods in single-cell sequencing predominantly concentrate on identifying cell clusters or delineating the principal progression line within the data. Spatial transcriptomics analysis primarily focuses on clustering and identifying spatially variable genes. While these methods effectively capture the primary features of the data, they may overlook more subtle processes, potentially involving specific subsets of samples. Also, widely employed techniques do not detect the regulatory activity of microRNAs, which play a crucial role in governing the expression of mRNAs and lncRNAs.

In this talk, I will introduce SPIRAL, an algorithm grounded in Gaussian statistics that is adept at identifying all statistically significant biological processes in single-cell, bulk, and spatial transcriptomics data. I will also present miTEA-HiRes, a method designed to facilitate the evaluation of microRNA activity at a high resolution. Lastly, I will speak about our work in detecting somatic point mutations in bulk RNA-seq samples.