Doan Nam Long Vu

@fnruji316625 @frankniujc Cc the first author @akatief48

A really interesting paper on representation geometry in LLMs written by my friend @frankniujc : “Hypothesis-Driven Feature Manifold Analysis in LLMs via SMDS” proposes a model-agnostic way to test geometric hypotheses about latent representations instead of assuming everything is just linear directions. They find that different concepts naturally form different structures like circles, lines, clusters, and that these manifolds remain surprisingly stable across model families/sizes while also dynamically reshaping with context. Very cool bridge between mechanistic interpretability and representation geometry. 🔥 Especially liked the framing that reasoning may operate over structured manifolds rather than isolated features. Paper: openreview.net/pdf?id=vCKZ40Y… Code: github.com/UKPLab/tmlr202… #LLM #MechanisticInterpretability #AIResearch #RepresentationLearning #TMLR #Interpretability #DeepLearning

It's a pleasure to work on this project 🙏

So happy to meet so many amazing people at #EACL2024 in 🇲🇹! We presented our framework for differentially private machine translation with MSc student @doannamlongvu, find out more about our paper "DP-NMT: Scalable Differentially-Private Machine Translation" below ⬇️

It seems that we won an outstanding paper award @coling2022 😊 This work was the result of @doannamlongvu bachelor thesis! and a great collaboration with @egere14 Looking forward to Doan’s future success🌺

Embedding-based eval metrics show high correlations with human eval But they are evaluated on benchmarks from similar domains as those of pretraining embeddings How to make them more robust on new/noisy domains with more unk tokens? @coling2022 led by @doannamlongvu with @egere14