Group-Theoretic Self-Supervised Representation Learning (Lecture)
On Jan 7, 2024, BIT’s “Hundred Lectures” will feature Assistant Professor Hanwang Zhang presenting his group‑theoretic self‑supervised representation learning work, including the IP‑IRM method that iteratively partitions data and applies invariant risk minimization to achieve fully disentangled visual features, with the session streamed via Tencent Meeting.
On January 7, 2024, from 15:00 to 17:00, the Beijing Institute of Technology “Hundred Lectures” series will host a lecture titled “Group-Theoretic Self‑Supervised Representation Learning” presented by Assistant Professor Hanwang Zhang from Nanyang Technological University.
Professor Zhang’s research spans computer vision, natural language processing, and causal inference. His talk introduces a group‑theoretic formulation of “good” visual representations, critiques existing self‑supervised learning (SSL) methods for capturing only low‑level augmentations, and proposes the Iterative Partition‑based Invariant Risk Minimization (IP‑IRM) algorithm to achieve full‑semantic disentanglement.
The abstract explains that IP‑IRM iteratively partitions training samples according to entangled semantic group actions, then applies invariant risk minimization to learn subset‑invariant similarities, guaranteeing disentanglement of the corresponding semantics. The method is proven to converge to a fully disentangled representation and demonstrates effectiveness on several SSL benchmarks.
The lecture will be held via Tencent Meeting (ID: 873‑681‑984) with a QR‑code provided for easy joining.
DataFunTalk
Dedicated to sharing and discussing big data and AI technology applications, aiming to empower a million data scientists. Regularly hosts live tech talks and curates articles on big data, recommendation/search algorithms, advertising algorithms, NLP, intelligent risk control, autonomous driving, and machine learning/deep learning.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.