Explaining Explainability: Understanding Concept Activation Vectors

Recent interpretability methods propose using concept-based explanations to translate the internal representations of deep learning models into a language that humans are familiar with: concepts. This requires understanding which concepts are present in the representation space of a neural network. One popular method for finding concepts is Concept Activation Vectors (CAVs), which are learnt using a probe dataset of concept exemplars. In this work, we investigate three properties of CAVs. CAVs may be: (1) inconsistent between layers, (2) entangled with different concepts, and (3) spatially dependent. Each property provides both challenges and opportunities in interpreting models. We introduce tools designed to detect the presence of these properties, provide insight into how they affect the derived explanations, and provide recommendations to minimise their impact. Understanding these properties can be used to our advantage. For example, we introduce spatially dependent CAVs to test if a model is translation invariant with respect to a specific concept and class. Our experiments are performed on ImageNet and a new synthetic dataset, Elements. Elements is designed to capture a known ground truth relationship between concepts and classes. We release this dataset to facilitate further research in understanding and evaluating interpretability methods.

Bayesian Preference Elicitation with Language Models

Aligning AI systems to users' interests requires understanding and incorporating humans' complex values and preferences. Recently, language models (LMs) have been used to gather information about the preferences of human users. This preference data can be used to fine-tune or guide other LMs and/or AI systems. However, LMs have been shown to struggle with crucial aspects of preference learning: quantifying uncertainty, modeling human mental states, and asking informative questions. These challenges have been addressed in other areas of machine learning, such as Bayesian Optimal Experimental Design (BOED), which focus on designing informative queries within a well-defined feature space. But these methods, in turn, are difficult to scale and apply to real-world problems where simply identifying the relevant features can be difficult. We introduce OPEN (Optimal Preference Elicitation with Natural language) a framework that uses BOED to guide the choice of informative questions and an LM to extract features and translate abstract BOED queries into natural language questions. By combining the flexibility of LMs with the rigor of BOED, OPEN can optimize the informativity of queries while remaining adaptable to real-world domains. In user studies, we find that OPEN outperforms existing LM- and BOED-based methods for preference elicitation.

Continual Learning via Sequential Function-Space Variational Inference

Sequential Bayesian inference over predictive functions is a natural framework for continual learning from streams of data. However, applying it to neural networks has proved challenging in practice. Addressing the drawbacks of existing techniques, we propose an optimization objective derived by formulating continual learning as sequential function-space variational inference. In contrast to existing methods that regularize neural network parameters directly, this objective allows parameters to vary widely during training, enabling better adaptation to new tasks. Compared to objectives that directly regularize neural network predictions, the proposed objective allows for more flexible variational distributions and more effective regularization. We demonstrate that, across a range of task sequences, neural networks trained via sequential function-space variational inference achieve better predictive accuracy than networks trained with related methods while depending less on maintaining a set of representative points from previous tasks.

Tractable Function-Space Variational Inference in Bayesian Neural Networks

Reliable predictive uncertainty estimation plays an important role in enabling the deployment of neural networks to safety-critical settings. A popular approach for estimating the predictive uncertainty of neural networks is to define a prior distribution over the network parameters, infer an approximate posterior distribution, and use it to make stochastic predictions. However, explicit inference over neural network parameters makes it difficult to incorporate meaningful prior information about the data-generating process into the model. In this paper, we pursue an alternative approach. Recognizing that the primary object of interest in most settings is the distribution over functions induced by the posterior distribution over neural network parameters, we frame Bayesian inference in neural networks explicitly as inferring a posterior distribution over functions and propose a scalable function-space variational inference method that allows incorporating prior information and results in reliable predictive uncertainty estimates. We show that the proposed method leads to state-of-the-art uncertainty estimation and predictive performance on a range of prediction tasks and demonstrate that it performs well on a challenging safety-critical medical diagnosis task in which reliable uncertainty estimation is essential.

Can Active Sampling Reduce Causal Confusion in Offline Reinforcement Learning?

Causal confusion is a phenomenon where an agent learns a policy that reflects imperfect spurious correlations in the data. Such a policy may falsely appear to be optimal during training if most of the training data contain such spurious correlations. This phenomenon is particularly pronounced in domains such as robotics, with potentially large gaps between the open- and closed-loop performance of an agent. In such settings, causally confused models may appear to perform well according to open-loop metrics during training but fail catastrophically when deployed in the real world. In this paper, we study causal confusion in offline reinforcement learning. We investigate whether selectively sampling appropriate points from a dataset of demonstrations may enable offline reinforcement learning agents to disambiguate the underlying causal mechanisms of the environment, alleviate causal confusion in offline reinforcement learning, and produce a safer model for deployment. To answer this question, we consider a set of tailored offline reinforcement learning datasets that exhibit causal ambiguity and assess the ability of active sampling techniques to reduce causal confusion at evaluation. We provide empirical evidence that uniform and active sampling techniques are able to consistently reduce causal confusion as training progresses and that active sampling is able to do so significantly more efficiently than uniform sampling.

DiscoBAX: Discovery of Optimal Intervention Sets in Genomic Experiment Design

The discovery of therapeutics to treat genetically-driven pathologies relies on identifying genes involved in the underlying disease mechanisms. Existing approaches search over the billions of potential interventions to maximize the expected influence on the target phenotype. However, to reduce the risk of failure in future stages of trials, practical experiment design aims to find a set of interventions that maximally change a target phenotype via diverse mechanisms. We propose DiscoBAX, a sample-efficient method for maximizing the rate of significant discoveries per experiment while simultaneously probing for a wide range of diverse mechanisms during a genomic experiment campaign. We provide theoretical guarantees of approximate optimality under standard assumptions, and conduct a comprehensive experimental evaluation covering both synthetic as well as real-world experimental design tasks. DiscoBAX outperforms existing state-of-the-art methods for experimental design, selecting effective and diverse perturbations in biological systems.

Form follows Function: Text-to-Text Conditional Graph Generation based on Functional Requirements

This work focuses on the novel problem setting of generating graphs conditioned on a description of the graph's functional requirements in a downstream task. We pose the problem as a text-to-text generation problem and focus on the approach of fine-tuning a pretrained large language model (LLM) to generate graphs. We propose an inductive bias which incorporates information about the structure of the graph into the LLM's generation process by incorporating message passing layers into an LLM's architecture. To evaluate our proposed method, we design a novel set of experiments using publicly available and widely studied molecule and knowledge graph data sets. Results suggest our proposed approach generates graphs which more closely meet the requested functional requirements, outperforming baselines developed on similar tasks by a statistically significant margin.

How to Catch an AI Liar: Lie Detection in Black-Box LLMs by Asking Unrelated Questions

Large language models (LLMs) can "lie", which we define as outputting false statements despite "knowing" the truth in a demonstrable sense. LLMs might "lie", for example, when instructed to output misinformation. Here, we develop a simple lie detector that requires neither access to the LLM's activations (black-box) nor ground-truth knowledge of the fact in question. The detector works by asking a predefined set of unrelated follow-up questions after a suspected lie, and feeding the LLM's yes/no answers into a logistic regression classifier. Despite its simplicity, this lie detector is highly accurate and surprisingly general. When trained on examples from a single setting -- prompting GPT-3.5 to lie about factual questions -- the detector generalises out-of-distribution to (1) other LLM architectures, (2) LLMs fine-tuned to lie, (3) sycophantic lies, and (4) lies emerging in real-life scenarios such as sales. These results indicate that LLMs have distinctive lie-related behavioural patterns, consistent across architectures and contexts, which could enable general-purpose lie detection.

Fine-tuning can cripple your foundation model; preserving features may be the solution

Pre-trained foundation models, owing primarily to their enormous capacity and exposure to vast amount of training data scraped from the internet, enjoy the advantage of storing knowledge about plenty of real-world concepts. Such models are typically fine-tuned on downstream datasets to produce remarkable state-of-the-art performances. While various fine-tuning methods have been devised and are shown to be highly effective, we observe that a fine-tuned model's ability to recognize concepts on tasks $\textit{different}$ from the downstream one is reduced significantly compared to its pre-trained counterpart. This is clearly undesirable as a huge amount of time and money went into learning those very concepts in the first place. We call this undesirable phenomenon "concept forgetting" and via experiments show that most end-to-end fine-tuning approaches suffer heavily from this side effect. To this end, we also propose a rather simple fix to this problem by designing a method called LDIFS (short for $\ell_2$ distance in feature space) that simply preserves the features of the original foundation model during fine-tuning. We show that LDIFS significantly reduces concept forgetting without having noticeable impact on the downstream task performance.

In-Context Learning in Large Language Models Learns Label Relationships but Is Not Conventional Learning

The performance of Large Language Models (LLMs) on downstream tasks often improves significantly when including examples of the input-label relationship in the context. However, there is currently no consensus about how this in-context learning (ICL) ability of LLMs works: for example, while Xie et al. (2021) liken ICL to a general-purpose learning algorithm, Min et al. (2022b) argue ICL does not even learn label relationships from in-context examples. In this paper, we study (1) how labels of in-context examples affect predictions, (2) how label relationships learned during pre-training interact with input-label examples provided in-context, and (3) how ICL aggregates label information across in-context examples. Our findings suggests LLMs usually incorporate information from in-context labels, but that pre-training and in-context label relationships are treated differently, and that the model does not consider all in-context information equally. Our results give insights into understanding and aligning LLM behavior.