Detecting and segmenting small objects, such as lung nodules and tumor lesions, remains a critical challenge in image analysis. These objects often occupy less than 0.1% of an image, making traditional transformer architectures inefficient and prone to performance degradation due to redundant attention computations on irrelevant regions. Existing sparse attention mechanisms rely on rigid hierarchical structures, which are poorly suited for detecting small, variable, and uncertain object locations. In this paper, we propose BoltzFormer, a novel transformer-based architecture designed to address these challenges through dynamic sparse attention. BoltzFormer identifies and focuses attention on relevant areas by modeling uncertainty using a Boltzmann distribution with an annealing schedule. Initially, a higher temperature allows broader area sampling in early layers, when object location uncertainty is greatest. As the temperature decreases in later layers, attention becomes more focused, enhancing efficiency and accuracy. BoltzFormer seamlessly integrates into existing transformer architectures via a modular Boltzmann attention sampling mechanism. Comprehensive evaluations on benchmark datasets demonstrate that BoltzFormer significantly improves segmentation performance for small objects while reducing attention computation by an order of magnitude compared to previous state-of-the-art methods.

Training and evaluating language models increasingly requires the construction of meta-datasets -- diverse collections of curated data with clear provenance. Natural language prompting has recently lead to improved zero-shot generalization by transforming existing, supervised datasets into a variety of novel instruction tuning tasks, highlighting the benefits of meta-dataset curation. While successful in general-domain text, translating these data-centric approaches to biomedical language modeling remains challenging, as labeled biomedical datasets are significantly underrepresented in popular data hubs. To address this challenge, we introduce BigBio a community library of 126+ biomedical NLP datasets, currently covering 13 task categories and 10+ languages. BigBio facilitates reproducible meta-dataset curation via programmatic access to datasets and their metadata, and is compatible with current platforms for prompt engineering and end-to-end few/zero shot language model evaluation. We discuss our process for task schema harmonization, data auditing, contribution guidelines, and outline two illustrative use cases: zero-shot evaluation of biomedical prompts and large-scale, multi-task learning. BigBio is an ongoing community effort and is available at https://github.com/bigscience-workshop/biomedical

Many biological studies involve either (i) manipulating some aspect of a cell or its environment and then simultaneously measuring the effect on thousands of genes, or (ii) systematically manipulating each gene and then measuring the effect on some response of interest. A common challenge that arises in these studies is to explain how genes identified as relevant in the given experiment are organized into a subnetwork that accounts for the response of interest. The task of inferring a subnetwork is typically dependent on the information available in publicly available, structured databases, which suffer from incompleteness. However, a wealth of potentially relevant information resides in the scientific literature, such as information about genes associated with certain concepts of interest, as well as interactions that occur among various biological entities. We contend that by exploiting this information, we can improve the explanatory power and accuracy of subnetwork inference in multiple applications. Here we propose and investigate several ways in which information extracted from the scientific literature can be used to augment subnetwork inference. We show that we can use literature-extracted information to (i) augment the set of entities identified as being relevant in a subnetwork inference task, (ii) augment the set of interactions used in the process, and (iii) support targeted browsing of a large inferred subnetwork by identifying entities and interactions that are closely related to concepts of interest. We use this approach to uncover the pathways involved in interactions between a virus and a host cell, and the pathways that are regulated by a transcription factor associated with breast cancer. Our experimental results demonstrate that these approaches can provide more accurate and more interpretable subnetworks. Integer program code, background network data, and pathfinding code are available at https://github.com/Craven-Biostat-Lab/subnetwork_inference.