Large language models (LLMs) encode vast amounts of world knowledge. However, since these models are trained on large swaths of internet data, they are at risk of inordinately capturing information about dominant groups. This imbalance can propagate into generated language. In this work, we study and operationalise a form of geographical erasure, wherein language models underpredict certain countries. We demonstrate consistent instances of erasure across a range of LLMs. We discover that erasure strongly correlates with low frequencies of country mentions in the training corpus. Lastly, we mitigate erasure by finetuning using a custom objective.
Off-policy evaluation (OPE) methods allow us to compute the expected reward of a policy by using the logged data collected by a different policy. OPE is a viable alternative to running expensive online A/B tests: it can speed up the development of new policies, and reduces the risk of exposing customers to suboptimal treatments. However, when the number of actions is large, or certain actions are under-explored by the logging policy, existing estimators based on inverse-propensity scoring (IPS) can have a high or even infinite variance. Saito and Joachims (arXiv:2202.06317v2 [cs.LG]) propose marginalized IPS (MIPS) that uses action embeddings instead, which reduces the variance of IPS in large action spaces. MIPS assumes that good action embeddings can be defined by the practitioner, which is difficult to do in many real-world applications. In this work, we explore learning action embeddings from logged data. In particular, we use intermediate outputs of a trained reward model to define action embeddings for MIPS. This approach extends MIPS to more applications, and in our experiments improves upon MIPS with pre-defined embeddings, as well as standard baselines, both on synthetic and real-world data. Our method does not make assumptions about the reward model class, and supports using additional action information to further improve the estimates. The proposed approach presents an appealing alternative to DR for combining the low variance of DM with the low bias of IPS.
Many state-of-the-art hyperparameter optimization (HPO) algorithms rely on model-based optimizers that learn surrogate models of the target function to guide the search. Gaussian processes are the de facto surrogate model due to their ability to capture uncertainty but they make strong assumptions about the observation noise, which might not be warranted in practice. In this work, we propose to leverage conformalized quantile regression which makes minimal assumptions about the observation noise and, as a result, models the target function in a more realistic and robust fashion which translates to quicker HPO convergence on empirical benchmarks. To apply our method in a multi-fidelity setting, we propose a simple, yet effective, technique that aggregates observed results across different resource levels and outperforms conventional methods across many empirical tasks.
Miscalibration-the mismatch between predicted probability and the true correctness likelihood-has been frequently identified in modern deep neural networks. Recent work in the field aims to address this problem by training calibrated models directly by optimizing a proxy of the calibration error alongside the conventional objective. Recently, Meta-Calibration (MC) showed the effectiveness of using meta-learning for learning better calibrated models. In this work, we extend MC with two main components: (1) gamma network (gamma-net), a meta network to learn a sample-wise gamma at a continuous space for focal loss for optimizing backbone network; (2) smooth expected calibration error (SECE), a Gaussian-kernel based unbiased and differentiable ECE which aims to smoothly optimizing gamma-net. The proposed method regularizes neural network towards better calibration meanwhile retain predictive performance. Our experiments show that (a) learning sample-wise gamma at continuous space can effectively perform calibration; (b) SECE smoothly optimise gamma-net towards better robustness to binning schemes; (c) the combination of gamma-net and SECE achieve the best calibration performance across various calibration metrics and retain very competitive predictive performance as compared to multiple recently proposed methods on three datasets.