talks

Download

talks@lists.cs.princeton.edu

November 2025

2 participants
22 discussions

Dan Friedman will present his FPO "Algorithmic Interpretability for Language Models" on Tuesday, November 11, 2025 at 1:00 PM in AI Lab Room 274 and Zoom.
by CS Grad Department 04 Nov '25

04 Nov '25

Dan Friedman will present his FPO " Algorithmic Interpretability for Language Models " on Tuesday, November 11, 2025 at 1:00 PM in AI Lab Room 274 and Zoom. Location: Zoom link: https://www.google.com/url?q=https%3A%2F%2Fprinceton.zoom.us%2Fj%2F86367410… The members of Dan ’s committee are as follows: Examiners: Danqi Chen (Adviser), Tom Griffiths , Adji Bousso Dieng Readers: Sanjeev Arora , Alexander Rush (Cornell University) A copy of his thesis is available upon request. Please email gradinfo(a)cs.princeton.edu if you would like a copy of the thesis. Everyone is invited to attend his talk. Abstract follows below: Modern natural language processing systems are black boxes. As a result, these models are fundamentally difficult to audit, debug, or trust. Standard post-hoc interpretability methods offer partial insight into model behavior but do not provide complete or faithful descriptions of the underlying computation. The aim of my research is to achieve an “algorithm-level” understanding of language processing systems, by aligning neural models with human-readable, rule-based programs. This thesis is organized into two parts. The first part involves developing intrinsically interpretable methods—machine learning models that are designed from the outset to be easily converted into symbolic form. I will present a method for finding interpretable features in text data sets using probabilistic grammars, which can recover “shortcuts” in NLP classification datasets—shallow patterns in the training data that are correlated with the target labels but do not generalize reliably. Next, I will introduce my work on developing transformers that are mechanistically interpretable by design, which can be trained using gradient-based optimization and then automatically converted into discrete, human-readable programs. These Transformer Programs are capable of learning effective solutions for a variety of tasks, and can be analyzed using standard code-analysis tools. The second part of the thesis involves aligning existing black-box models with symbolic rules, either by theoretical analysis or by extracting rule-based descriptions of their behavior. This includes a theoretical and empirical analysis of how transformers can represent classic, rule-based chatbot algorithms like ELIZA, and a method for extracting rule-based explanations of attention features using Sparse Autoencoders. I will conclude by outlining a roadmap for future research towards the goal of algorithm-level understanding of neural language models.

1 0

PPH seminar: Digital Phenotyping for Precision Mental Health - Friday, Nov 7th at 12pm
by Emily C. Lawrence 03 Nov '25

03 Nov '25

Please join the Princeton Precision Health (PPH) Initiative for the next talk of our Fall 2025 semester series on Friday, November 7th, at 12:00 pm, at 252 Nassau Street. Dr. Nelson Freimer is Director of UCLA’s Depression Grand Challenge, a campus-wide initiative that aims to cut the burden of depression in half by 2050. He will present a seminar on, Digital Phenotyping for Precision Mental Health. Widespread application of digital phenotyping – assessments using data passively obtained from smartphones and wearables – could transform mental health research and clinical care by providing nearly continuous, quantitative, and objective information across broad physiological, behavioral, and emotional domains. The UCLA-Apple Digital Mental Health Study (DMHS) is the largest and longest digital mental health study conducted to date, having collected up to12 months of sensor data from phones and watches (resulting in >900 phenotypic features) in a sample of >4,000 participants, diverse by age, sex at birth, ethnicity, and depression symptom severity. Analyses of DMHS data identify highly specific associations between depression and anxiety symptoms and sensor domains. Additionally, the results suggest that longitudinal digital phenotyping will enable dissection of heterogenous diagnostic categories, a key step towards precision mental health. Lunch will be provided. Getting to the seminar space currently requires that you climb a set of stairs. If an accommodation is needed, please contact PPH in advance at: [ mailto:PrincetonPPH@princeton.edu | PrincetonPPH(a)princeton.edu ]

1 0