Colloquia and Seminars

As LLMs generate increasingly long outputs, effective uncertainty estimation must identify errors at fine-grained levels rather than discard entire responses. While such methods exist, evaluating uncertainty at any resolution (token to an entire generation) is challenging and highly sensitive to label imperfections, making zero-noise benchmarks essential; yet, long-form generation benchmarks tend to rely on fallible labels rather than deterministic ground truth.

We introduce Single-answer Atomic Long-form Target (SALT), a benchmark of six procedurally generated tasks with single deterministic long textual ground truths, enabling unit-level evaluation of correctness, calibration, and ranking without external judges. Equipped with SALT, our analysis of 50+ LLMs reveals key insights: We identify which confidence functions dominate each uncertainty aspect and show that effective ranking benefits more from coarser evaluation resolutions; SALT further facilitates precise calibration tracking throughout generation, revealing a divergence in the accuracy–calibration relationship, with high- and low-performing models exhibiting degradation ($\rho=0.87$) and improvement ($\rho=-0.92$).

Finally, we demonstrate that reasoning, via Chain-of-Thought prompting or internalized through training, introduces a trade-off, improving accuracy while degrading confidence ranking. These findings directly impact risk-critical applications requiring reliable error identification and mitigation.

Protein sequences are abundant in repeating segments, both as exact copies and as approximate segments with mutations. These repeats are important for protein structure and function, motivating decades of algorithmic work on repeat identification. Recent work has shown that protein language models (PLMs) identify repeats, by examining their behavior in masked-token prediction.

To elucidate their internal mechanisms, we investigate how PLMs detect both exact and approximate repeats. We find that the mechanism for approximate repeats functionally subsumes that of exact repeats.

We then characterize this mechanism, revealing two main stages: PLMs first build feature representations using both general positional attention heads and biologically specialized components, such as neurons that encode amino-acid similarity. Then, induction heads attend to aligned tokens across repeated segments, promoting the correct answer.

Our results reveal how PLMs solve this biological task by combining language-based pattern matching with specialized biological knowledge, thereby establishing a basis for studying more complex evolutionary processes in PLMs.

The study of spectral graph determination is a central and fascinating topic in spectral graph theory and algebraic combinatorics. This area investigates the spectral characterization of various classes of graphs, develops methods for constructing and distinguishing cospectral nonisomorphic graphs, and analyzes the conditions under which the spectrum of a graph uniquely determines its structure. In the first part of the seminar, we present both classical results and recent advances in spectral graph determination.

The study of graph symmetries and different notions of transitivity is also of fundamental interest in algebraic graph theory. In the second part of the talk, we examine transitivity properties of Gilbert graphs and their complements, and discuss the main ideas underlying these results.

Recent advances in computer vision, foundation models, and transformer architectures have transformed computational pathology, enabling deep learning systems to extract clinically actionable information directly from digitized tissue slides. This seminar explores how these technologies come together in modern digital pathology frameworks, and presents two studies demonstrating their clinical impact.

The first study addresses a critical diagnostic gap in low-resource settings, showing that convolutional neural networks applied to Giemsa-stained bone marrow aspirates can predict B/T-cell lineage and ETV6–RUNX1 translocation status in pediatric acute lymphoblastic leukemia — tasks that traditionally require expensive molecular assays unavailable in many parts of the world.

The second study tackles overtreatment in breast cancer. The TAILORx trial established that adjuvant chemotherapy can be spared for postmenopausal HR+/HER2− node-negative breast cancer patients with a 21-gene Recurrence Score (RS) of 11–25. However, among premenopausal women with RS 16–25, a small benefit from chemotherapy could not be ruled out. Consequently, guidelines suggest considering chemotherapy for this population, creating a therapeutic dilemma and leading to widespread overtreatment of patients who may not benefit from chemotherapy. Using deep survival analysis on H&E whole-slide images, we identify which women in this group truly benefit from adjuvant chemotherapy. Our model stratifies 76% of this population as low-risk, for whom chemotherapy can be safely omitted, while correctly identifying the high-risk subset that benefits from treatment.

Together, these works illustrate how digital pathology can democratize access to precision diagnostics and enable more personalized, less toxic cancer care.