Discovering Preference Optimization Algorithms with and for Large Language Models

Offline preference optimization is a key method for enhancing and controlling the quality of Large Language Model (LLM) outputs. Typically, preference optimization is approached as an offline supervised learning task using manually-crafted convex loss functions. While these methods are based on theoretical insights, they are inherently constrained by human creativity, so the large search space of possible loss functions remains under explored. We address this by performing LLM-driven objective discovery to automatically discover new state-of-the-art preference optimization algorithms without (expert) human intervention. Specifically, we iteratively prompt an LLM to propose and implement new preference optimization loss functions based on previously-evaluated performance metrics. This process leads to the discovery of previously-unknown and performant preference optimization algorithms. The best performing of these we call Discovered Preference Optimization (DiscoPOP), a novel algorithm that adaptively blends logistic and exponential losses. Experiments demonstrate the state-of-the-art performance of DiscoPOP and its successful transfer to held-out tasks.

This Reads Like That: Deep Learning for Interpretable Natural Language Processing

Prototype learning, a popular machine learning method designed for inherently interpretable decisions, leverages similarities to learned prototypes for classifying new data. While it is mainly applied in computer vision, in this work, we build upon prior research and further explore the extension of prototypical networks to natural language processing. We introduce a learned weighted similarity measure that enhances the similarity computation by focusing on informative dimensions of pre-trained sentence embeddings. Additionally, we propose a post-hoc explainability mechanism that extracts prediction-relevant words from both the prototype and input sentences. Finally, we empirically demonstrate that our proposed method not only improves predictive performance on the AG News and RT Polarity datasets over a previous prototype-based approach, but also improves the faithfulness of explanations compared to rationale-based recurrent convolutions.

Natural Language Processing Methods to Identify Oncology Patients at High Risk for Acute Care with Clinical Notes

Clinical notes are an essential component of a health record. This paper evaluates how natural language processing (NLP) can be used to identify the risk of acute care use (ACU) in oncology patients, once chemotherapy starts. Risk prediction using structured health data (SHD) is now standard, but predictions using free-text formats are complex. This paper explores the use of free-text notes for the prediction of ACU instead of SHD. Deep Learning models were compared to manually engineered language features. Results show that SHD models minimally outperform NLP models; an l1-penalised logistic regression with SHD achieved a C-statistic of 0.748 (95%-CI: 0.735, 0.762), while the same model with language features achieved 0.730 (95%-CI: 0.717, 0.745) and a transformer-based model achieved 0.702 (95%-CI: 0.688, 0.717). This paper shows how language models can be used in clinical applications and underlines how risk bias is different for diverse patient groups, even using only free-text data.

This Looks Like That Does it? Shortcomings of Latent Space Prototype Interpretability in Deep Networks

Deep neural networks that yield human interpretable decisions by architectural design have lately become an increasingly popular alternative to post hoc interpretation of traditional black-box models. Among these networks, the arguably most widespread approach is so-called prototype learning, where similarities to learned latent prototypes serve as the basis of classifying an unseen data point. In this work, we point to an important shortcoming of such approaches. Namely, there is a semantic gap between similarity in latent space and similarity in input space, which can corrupt interpretability. We design two experiments that exemplify this issue on the so-called ProtoPNet. Specifically, we find that this network's interpretability mechanism can be led astray by intentionally crafted or even JPEG compression artefacts, which can produce incomprehensible decisions. We argue that practitioners ought to have this shortcoming in mind when deploying prototype-based models in practice.