In [52], the choice of their parametrization (i.e., shifting and rescaling) is motivated by the example

This paper investigates the knowledge of language models from the perspective of Bayesian epistemology. We explore how language models adjust their confidence and responses when presented with evidence with varying levels of informativeness and reliability. To study these properties, we create a dataset with various types of evidence and analyze language models' responses and confidence using verbalized confidence, token probability, and sampling. We observed that language models do not consistently follow Bayesian epistemology: language models follow the Bayesian confirmation assumption well with true evidence but fail to adhere to other Bayesian assumptions when encountering different evidence types. Also, we demonstrated that language models can exhibit high confidence when given strong evidence, but this does not always guarantee high accuracy. Our analysis also reveals that language models are biased toward golden evidence and show varying performance depending on the degree of irrelevance, helping explain why they deviate from Bayesian assumptions.

A model is considered well-calibrated when its probability estimate aligns with the actual likelihood of the output being correct. Calibrating language models (LMs) is crucial, as it plays a vital role in detecting and mitigating hallucinations, a common issue of LMs, as well as building more trustworthy models. Yet, popular neural model calibration techniques are not well-suited for LMs due to their lack of flexibility in discerning answer correctness and their high computational costs. For instance, post-processing methods like temperature scaling are often unable to reorder the candidate generations. Moreover, training-based methods require finetuning the entire model, which is impractical due to the increasing sizes of modern LMs. In this paper, we present LitCab, a lightweight calibration mechanism consisting of a single linear layer taking the input text representation and manipulateing the LM output logits. LitCab improves model calibration by only adding < 2% of the original model parameters. For evaluation, we construct CaT, a benchmark consisting of 7 text generation tasks, covering responses ranging from short phrases to paragraphs. We test LitCab with Llama2-7B, where it improves calibration across all tasks, by reducing the average ECE score by 20%. We further conduct a comprehensive evaluation with 7 popular open-sourced LMs from GPT and LLaMA families, yielding the following key findings: (1) Larger models within the same family exhibit better calibration on tasks with short generation tasks, but not necessarily for longer ones. (2) GPT-family models show superior calibration compared to LLaMA, Llama2 and Vicuna models despite having much fewer parameters. (3) Finetuning pretrained model (e.g., LLaMA) with samples of limited purpose (e.g., conversations) may lead to worse calibration, highlighting the importance of finetuning setups for calibrating LMs.

Social Networking Sites (SNS) are one of the most important ways of communication. In particular, microblogging sites are being used as analysis avenues due to their peculiarities (promptness, short texts...). There are countless researches that use SNS in novel manners, but machine learning (ML) has focused mainly in classification performance rather than interpretability and/or other goodness metrics. Thus, state-of-the-art models are black boxes that should not be used to solve problems that may have a social impact. When the problem requires transparency, it is necessary to build interpretable pipelines. Arguably, the most decisive component in the pipeline is the classifier, but it is not the only thing that we need to consider. Despite that the classifier may be interpretable, resulting models are too complex to be considered comprehensible, making it impossible for humans to comprehend the actual decisions. The purpose of this paper is to present a feature selection mechanism (the first step in the pipeline) that is able to improve comprehensibility by using less but more meaningful features while achieving a good performance in microblogging contexts where interpretability is mandatory. Moreover, we present a ranking method to evaluate features in terms of statistical relevance and bias. We conducted exhaustive tests with five different datasets in order to evaluate classification performance, generalisation capacity and actual interpretability of the model. Our results shows that our proposal is better and, by far, the most stable in terms of accuracy, generalisation and comprehensibility.

Ensembles over neural network weights trained from different random initialization, known as deep ensembles, achieve state-of-the-art accuracy and calibration. The recently introduced batch ensembles provide a drop-in replacement that is more parameter efficient. In this paper, we design ensembles not only over weights, but over hyperparameters to improve the state of the art in both settings. For best performance independent of budget, we propose hyper-deep ensembles, a simple procedure that involves a random search over different hyperparameters, themselves stratified across multiple random initializations. Its strong performance highlights the benefit of combining models with both weight and hyperparameter diversity. We further propose a parameter efficient version, hyper-batch ensembles, which builds on the layer structure of batch ensembles and self-tuning networks. The computational and memory costs of our method are notably lower than typical ensembles. On image classification tasks, with MLP, LeNet, ResNet 20 and Wide ResNet 28-10 architectures, we improve upon both deep and batch ensembles.

In this paper, we study the problem of deriving fast and accurate classification algorithms with uncertainty quantification. Gaussian process classification provides a principled approach, but the corresponding computational burden is hardly sustainable in large-scale problems and devising efficient alternatives is a challenge. In this work, we investigate if and how Gaussian process regression directly applied to the classification labels can be used to tackle this question. While in this case training time is remarkably faster, predictions need be calibrated for classification and uncertainty estimation. To this aim, we propose a novel approach based on interpreting the labels as the output of a Dirichlet distribution. Extensive experimental results show that the proposed approach provides essentially the same accuracy and uncertainty quantification of Gaussian process classification while requiring only a fraction of computational resources.

A set of probabilistic predictions is well calibrated if the events that are predicted to occur with probability p do in fact occur about p fraction of the time. Well calibrated predictions are particularly important when machine learning models are used in decision analysis. This paper presents two new non-parametric methods for calibrating outputs of binary classification models: a method based on the Bayes optimal selection and a method based on the Bayesian model averaging. The advantage of these methods is that they are independent of the algorithm used to learn a predictive model, and they can be applied in a post-processing step, after the model is learned. This makes them applicable to a wide variety of machine learning models and methods. These calibration methods, as well as other methods, are tested on a variety of datasets in terms of both discrimination and calibration performance. The results show the methods either outperform or are comparable in performance to the state-of-the-art calibration methods.