Introducing a new blog post on interpretable data fusion for distributed learning via gradient matching. Unlike traditional methods, our approach offers human interpretability and maintains privacy. Check it out at: bit.ly/3UQeCok #machinelearning #datascience

I'll present our #ICLR2024 poster #167 'Non-negative Contrastive Learning' tmr 10:45am. A simple (one-line) technique that gives much better feature interpretability while improving (at least maintaining) performance.

Welcome to drop by our poster and have a chat!

Zimmermann+, openreview.net/forum?id=OZ7aI…
Mechanistic interpretabilityは，DNNを細かい単位に分解し，リバースエンジニアリングによりその内部の働きを解釈する試み．人間に各ユニットの挙動を予想させた時の正答率で解釈性スコアを測ると，データセットやモデルのサイズによる差は見られなかった．

[CL] A Primer on the Inner Workings of Transformer-based Language Models
arxiv.org/abs/2405.00208
- The paper provides a concise technical introduction to interpretability techniques used to analyze Transformer-based language models, focusing on the generative decoder-only…

Okay, first pass over KAN: Kolmogorov–Arnold Networks, it looks very interesting!

Interpretability of KAN model:
May be considered mostly as a safety issue these days, but it can also be used as a form of interaction between the user and a model, as this paper argues and I…

In his talk at #TAIS , Stan van Wingerden shared the discoveries of singular learning theory and how they pave the way for fresh prospects in interpretability, mechanistic anomaly detection, and the exploration of inductive biases. He elaborated on his vision for the field's…

*A Primer on the Inner Workings of Transformer LMs*
by Javier Ferrando Gabriele Sarti Arianna Bisazza Marta R. Costa-jussa

I was waiting for this! Cool comprehensive survey on interpretability methods for LLMs, with a focus on recent techniques (e.g., logit lens).

arxiv.org/abs/2405.00208

Just came across this paper introducing Kolmogorov-Arnold Networks (KANs): a breakthrough in deep learning that replaces fixed activation functions in MLPs with learnable activations on edges, achieving superior accuracy and interpretability compared to classic MLPs!

AI 2.0 ?

[1/4] Introducing “A Primer on the Inner Workings of Transformer-based Language Models”, a comprehensive survey on interpretability methods and the findings into the functioning of language models they have led to.

ArXiv: arxiv.org/pdf/2405.00208

Researchers at MIT Propose ‘MAIA’: An Artificial Intelligence System that Uses Neural Network Models to Automate Neural Model Understanding Tasks
A Multimodal Automated Interpretability Agent

If you're at #ICLR2024 , drop by poster
#263 to hear about pitfalls of using subspace activation patching for interpretability! Joint work with Georg Lange, Atticus Geiger, Neel Nanda

Ever wondered how to interpret your ML models? 🤔
We explain a powerful interpretability technique: Shapley Values – can be used to explain any model, including LLMs!
💻 We show simple code for how to use them and 📖 dive into the theory behind them.
📺 youtu.be/5-1lKFvV1i0

I need Trenton Bricken’s interpretability T-shirt

This amazing NSF workshop on New Horizons in Language Science will be streamed online!

My abstract on the topic What key future scientific opportunities lie at the interface between the study of human language and large language model development?

Interpretability techniques…

A warm welcome to Subhabrata Dutta, who has just started as a Postdoctoral Researcher at UKP Lab! 👋 Subhabrata joins us from LCS2 Lab, his areas of interest are Mechanistic Interpretability and #LLM Security.

Find out more about him on GitHub: subha0009.github.io

17/N Given our empirical results, we believe that KANs will be a useful model/tool for AI + Science due to their accuracy, parameter efficiency and interpretability. The usefulness of KANs for machine learning-related tasks is more speculative and left for future work.

'Could a Mechanistic Interpretability Researcher Understand the Linux Kernel?'

(journals.plos.org/ploscompbiol/a…)

Together with Yves Scherrer and Aleksandra Miletić I had a look at explainability/Interpretability of a machine learning approach in the forensic linguistic context. Our preprint is now out on arxiv: arxiv.org/abs/2404.18510

I joined the mechanistic interpretability team at Anthropic recently, and it's been really exciting. To any comp. neuroscience followers -- I highly recommend following this literature, or getting involved yourself! You can keep up with our research here: transformer-circuits.pub

Focus on interpretability to build trust in AI systems. Making your models transparent and understandable helps stakeholders see how decisions are made, boosting confidence in AI solutions. Clear, explainable AI is essential for widespread adoption. #TrustInAI #ExplainableAI