David Scott Krueger

Dmitrii Krasheninnikov

Usman Anwar

A key objective of interpretability research on large language models (LLMs) is to develop methods for robustly steering models toward desir… (voir plus)ed behaviors. To this end, two distinct approaches to interpretability -- ``bottom-up"and ``top-down"-- have been presented, but there has been little quantitative comparison between them. We present a case study comparing the effectiveness of representative vector steering methods from each branch: function vectors (FV; arXiv:2310.15213), as a bottom-up method, and in-context vectors (ICV; arXiv:2311.06668) as a top-down method. While both aim to capture compact representations of broad in-context learning tasks, we find they are effective only on specific types of tasks: ICVs outperform FVs in behavioral shifting, whereas FVs excel in tasks requiring more precision. We discuss the implications for future evaluations of steering methods and for further research into top-down and bottom-up steering given these findings.

2024-11-11

ArXiv (prépublication)

PoisonBench: Assessing Large Language Model Vulnerability to Data Poisoning

Tingchen Fu

Mrinank Sharma

Philip Torr

Shay B. Cohen

Fazl Barez

Preference learning is a central component for aligning current LLMs, but this process can be vulnerable to data poisoning attacks. To addre… (voir plus)ss this concern, we introduce PoisonBench, a benchmark for evaluating large language models' susceptibility to data poisoning during preference learning. Data poisoning attacks can manipulate large language model responses to include hidden malicious content or biases, potentially causing the model to generate harmful or unintended outputs while appearing to function normally. We deploy two distinct attack types across eight realistic scenarios, assessing 21 widely-used models. Our findings reveal concerning trends: (1) Scaling up parameter size does not inherently enhance resilience against poisoning attacks; (2) There exists a log-linear relationship between the effects of the attack and the data poison ratio; (3) The effect of data poisoning can generalize to extrapolated triggers that are not included in the poisoned data. These results expose weaknesses in current preference learning techniques, highlighting the urgent need for more robust defenses against malicious models and data manipulation.

2024-10-11

ArXiv (prépublication)

Input Space Mode Connectivity in Deep Neural Networks

Jakub Vrabel

Ori Shem-Ur

Yaron Oz

We extend the concept of loss landscape mode connectivity to the input space of deep neural networks. Initially studied in parameter space, … (voir plus)mode connectivity describes the existence of low-loss paths between solutions (loss minimizers) found via gradient descent. We present theoretical and empirical evidence of its presence in the input space of deep networks, thereby highlighting the broader nature of the phenomenon. We observe that different input images with similar predictions are generally connected, and for trained models, the path tends to be simple, with only a small deviation from being a linear path. We conjecture that input space mode connectivity in high-dimensional spaces is a geometric phenomenon, present even in untrained models, and can be explained by percolation theory. We exploit mode connectivity to obtain new insights about adversarial examples and show its potential for adversarial detection and interpretability.

2024-10-10

NeurIPS.cc/2024/Workshop/SciForDL (présentation orale)

Sparse Autoencoders Reveal Universal Feature Spaces Across Large Language Models

Michael Lan

Philip Torr

Austin Meek

Ashkan Khakzar

Fazl Barez

2024-10-09

ArXiv (prépublication)

Towards Interpreting Visual Information Processing in Vision-Language Models

Clement Neo

Luke Ong

Philip Torr

Mor Geva

Fazl Barez

2024-10-09

ArXiv (prépublication)

A Generative Model of Symmetry Transformations

James Urquhart Allingham

Bruno Mlodozeniec

Shreyas Padhy

Javier Antoran

Richard E. Turner

Eric Nalisnick

José Miguel Hernández-Lobato

Correctly capturing the symmetry transformations of data can lead to efficient models with strong generalization capabilities, though method… (voir plus)s incorporating symmetries often require prior knowledge. While recent advancements have been made in learning those symmetries directly from the dataset, most of this work has focused on the discriminative setting. In this paper, we take inspiration from group theoretic ideas to construct a generative model that explicitly aims to capture the data's approximate symmetries. This results in a model that, given a prespecified broad set of possible symmetries, learns to what extent, if at all, those symmetries are actually present. Our model can be seen as a generative process for data augmentation. We provide a simple algorithm for learning our generative model and empirically demonstrate its ability to capture symmetries under affine and color transformations, in an interpretable way. Combining our symmetry model with standard generative models results in higher marginal test-log-likelihoods and improved data efficiency.

2024-09-25

NeurIPS.cc/2024/Conference (poster)

Interpreting Learned Feedback Patterns in Large Language Models

Luke Marks

Amir Abdullah

Clement Neo

Rauno Arike

Philip Torr

Fazl Barez

2024-09-25

NeurIPS.cc/2024/Conference (poster)

Predicting Future Actions of Reinforcement Learning Agents

Stephen Chung

Scott Niekum

2024-09-25

NeurIPS.cc/2024/Conference (poster)

Stress-Testing Capability Elicitation With Password-Locked Models

Ryan Greenblatt

Fabien Roger

Dmitrii Krasheninnikov

2024-09-25

NeurIPS.cc/2024/Conference (poster)

Foundational Challenges in Assuring Alignment and Safety of Large Language Models

Usman Anwar

Abulhair Saparov

Javier Rando

Daniel Paleka

Miles Turpin

Peter Hase

Ekdeep Singh Lubana

Erik Jenner

Stephen Casper

Oliver Sourbut

Benjamin L. Edelman

Zhaowei Zhang

Mario Günther

Anton Korinek

Jose Hernandez-Orallo

Lewis Hammond

Eric J Bigelow

Alexander Pan

Lauro Langosco

Tomasz Korbak … (voir 22 de plus)

Heidi Chenyu Zhang

Ruiqi Zhong

Sean O hEigeartaigh

Gabriel Recchia

Giulio Corsi

Alan Chan

Markus Anderljung

Lilian Edwards

Aleksandar Petrov

Yoshua Bengio

Christian Schroeder de Witt

Danqi Chen

Sumeet Ramesh Motwani

Samuel Albanie

Tegan Maharaj

Jakob Nicolaus Foerster

Philip Torr

Florian Tramèr

He He

Atoosa Kasirzadeh

Yejin Choi

2024-09-02

TMLR (accepté)

Implicitly Bayesian Prediction Rules in Deep Learning

Bruno Mlodozeniec

Richard Turner

The Bayesian approach leads to coherent updates of predictions under new data, which makes adhering to Bayesian principles appealing in deci… (voir plus)sion-making contexts. Traditionally, integrating Bayesian principles into models like deep neural networks involves setting priors on parameters and approximating posteriors. This is done despite the fact that, typically, priors on parameters reflect any prior beliefs only insofar as they dictate function space behaviour. In this paper, we rethink this approach and consider what properties characterise a prediction rule as being Bayesian. Algorithms meeting such criteria can be deemed implicitly Bayesian — they make the same predictions as some Bayesian model, without explicitly manifesting priors and posteriors. We argue this might be a more fruitful approach towards integrating Bayesian principles into deep learning. In this paper, we propose how to measure how close a general prediction rule is to being implicitly Bayesian, and empirically evaluate multiple prediction strategies using our approach. We also show theoretically that agents relying on non-implicitly Bayesian prediction rules can be easily exploited in adversarial betting settings.

2024-07-29

Proceedings of the 6th Symposium on Advances in Approximate Bayesian Inference (publié)

proceedings.mlr.press

Implicit meta-learning may lead language models to trust more reliable sources

Dmitrii Krasheninnikov

Egor Krasheninnikov

Bruno Mlodozeniec

Tegan Maharaj

We demonstrate that large language models (LLMs) may learn indicators of document usefulness and modulate their updates accordingly. We intr… (voir plus)oduce random strings ("tags") as indicators of usefulness in a synthetic fine-tuning dataset. Fine-tuning on this dataset leads to **implicit meta-learning (IML)**: in further fine-tuning, the model updates to make more use of text that is tagged as useful. We perform a thorough empirical investigation of this phenomenon, finding (among other things) that (i) it occurs in both pretrained LLMs and those trained from scratch, as well as on a vision task, and (ii) larger models and smaller batch sizes tend to give more IML. We also use probing to examine how IML changes the way models store knowledge in their parameters. Finally, we reflect on what our results might imply about the capabilities, risks, and controllability of future AI systems.

2024-07-08

Proceedings of the 41st International Conference on Machine Learning (publié)

proceedings.mlr.press