Proper losses such as cross-entropy incentivize classifiers to produce class probabilities that are well-calibrated on the training data. Due to the generalization gap, these classifiers tend to become overconfident on the test data, mandating calibration methods such as temperature scaling. The focal loss is not proper, but training with it has been shown to often result in classifiers that are better calibrated on test data. Our first contribution is a simple explanation about why focal loss training often leads to better calibration than cross-entropy training. For this, we prove that focal loss can be decomposed into a confidence-raising transformation and a proper loss. This is why focal loss pushes the model to provide under-confident predictions on the training data, resulting in being better calibrated on the test data, due to the generalization gap. Secondly, we reveal a strong connection between temperature scaling and focal loss through its confidence-raising transformation, which we refer to as the focal calibration map. Thirdly, we propose focal temperature scaling - a new post-hoc calibration method combining focal calibration and temperature scaling. Our experiments on three image classification datasets demonstrate that focal temperature scaling outperforms standard temperature scaling.

This paper proposes an easy-to-use method for one-class classification: Repeated Element-wise Folding (REF). The algorithm consists of repeatedly standardizing and applying an element-wise folding operation on the one-class training data. Equivalent mappings are performed on unknown test items and the classification prediction is based on the item's distance to the origin of the final distribution. As all the included operations have linear time complexity, the proposed algorithm provides a linear-time alternative for the commonly used computationally much more demanding approaches. Furthermore, REF can avoid the challenges of hyperparameter setting in one-class classification by providing robust default settings. The experiments show that the proposed method can produce similar classification performance or even outperform the more complex algorithms on various benchmark datasets. Matlab codes for REF are publicly available at https://github.com/JenniRaitoharju/REF.

Recent advances in self-supervised learning enabled novel medical AI models, known as foundation models (FMs) that offer great potential for characterizing health from diverse biomedical data. Continuous glucose monitoring (CGM) provides rich, temporal data on glycemic patterns, but its full potential for predicting broader health outcomes remains underutilized. Here, we present GluFormer, a generative foundation model on biomedical temporal data based on a transformer architecture, and trained on over 10 million CGM measurements from 10,812 non-diabetic individuals. We tokenized the CGM training data and trained GluFormer using next token prediction in a generative, autoregressive manner. We demonstrate that GluFormer generalizes effectively to 15 different external datasets, including 4936 individuals across 5 different geographical regions, 6 different CGM devices, and several metabolic disorders, including normoglycemic, prediabetic, and diabetic populations, as well as those with gestational diabetes and obesity. GluFormer produces embeddings which outperform traditional CGM analysis tools, and achieves high Pearson correlations in predicting clinical parameters such as HbA1c, liver-related parameters, blood lipids, and sleep-related indices. Notably, GluFormer can also predict onset of future health outcomes even 4 years in advance. We also show that CGM embeddings from pre-intervention periods in Randomized Clinical Trials (RCTs) outperform other methods in predicting primary and secondary outcomes. When integrating dietary data into GluFormer, we show that the enhanced model can accurately generate CGM data based only on dietary intake data, simulate outcomes of dietary interventions, and predict individual responses to specific foods. Overall, we show that GluFormer accurately predicts health outcomes which generalize across different populations metabolic conditions.

Main Text Word Count: 4163 Figures Count: 3 Tables Count: 3 ABSTRACT Large Language Models (LLMs) have demonstrated significant potential in clinical applications through prompt engineering, which enables the generation of flexible and diverse clinical predictions. However, they pose challenges in producing prediction probabilities, which are essential for transparency and allowing clinicians to apply flexible probability thresholds in decision-making. While explicit prompt instructions can lead LLMs to provide prediction probability numbers through text generation, LLMs' limitations in numerical reasoning raise concerns about the reliability of these text-generated probabilities. To assess this reliability, we compared explicit probabilities derived from text generation to implicit probabilities calculated based on the likelihood of predicting the correct label token. Experimenting with six advanced open-source LLMs across five medical datasets, we found that the performance of explicit probabilities was consistently lower than implicit probabilities with respect to discrimination, precision, and recall. Moreover, these differences were enlarged on small LLMs and imbalanced datasets, emphasizing the need for cautious interpretation and applications, as well as further research into robust probability estimation methods for LLMs in clinical contexts. BACKGROUND Generating credible probability of prediction is crucial in clinical practice and medical research when applying artificial intelligence (AI) to healthcare. Accurate probability outputs significantly influence the adoption and effectiveness of AI in healthcare. Existing literature has utilized the generation of probability values directly through text generation by LLMs, this text-generated probability, or "explicit probability" is simple and flexible For example, for the question "Given the following lab reports: [lab report text]. Does the patient have COVID?

Detecting out-of-distribution (OOD) data is crucial in machine learning applications to mitigate the risk of model overconfidence, thereby enhancing the reliability and safety of deployed systems. The majority of existing OOD detection methods predominantly address uni-modal inputs, such as images or texts. In the context of multi-modal documents, there is a notable lack of extensive research on the performance of these methods, which have primarily been developed with a focus on computer vision tasks. We propose a novel methodology termed as attention head masking (AHM) for multi-modal OOD tasks in document classification systems. Our empirical results demonstrate that the proposed AHM method outperforms all state-of-the-art approaches and significantly decreases the false positive rate (FPR) compared to existing solutions up to 7.5\%. This methodology generalizes well to multi-modal data, such as documents, where visual and textual information are modeled under the same Transformer architecture. To address the scarcity of high-quality publicly available document datasets and encourage further research on OOD detection for documents, we introduce FinanceDocs, a new document AI dataset. Our code and dataset are publicly available.

Modeling the errors of a speech recognizer can help simulate errorful recognized speech data from plain text, which has proven useful for tasks like discriminative language modeling, improving robustness of NLP systems, where limited or even no audio data is available at train time. Previous work typically considered replicating behavior of GMM-HMM based systems, but the behavior of more modern posterior-based neural network acoustic models is not the same and requires adjustments to the error prediction model. In this work, we extend a prior phonetic confusion based model for predicting speech recognition errors in two ways: first, we introduce a sampling-based paradigm that better simulates the behavior of a posterior-based acoustic model. Second, we investigate replacing the confusion matrix with a sequence-to-sequence model in order to introduce context dependency into the prediction. We evaluate the error predictors in two ways: first by predicting the errors made by a Switchboard ASR system on unseen data (Fisher), and then using that same predictor to estimate the behavior of an unrelated cloud-based ASR system on a novel task. Sampling greatly improves predictive accuracy within a 100-guess paradigm, while the sequence model performs similarly to the confusion matrix.

In this paper, the causal bandit problem is investigated, in which the objective is to select an optimal sequence of interventions on nodes in a causal graph. It is assumed that the graph is governed by linear structural equations; it is further assumed that both the causal topology and the distribution of interventions are unknown. By exploiting the causal relationships between the nodes whose signals contribute to the reward, interventions are optimized. First, based on the difference between the two types of graph identification errors (false positives and negatives), a causal graph learning method is proposed, which strongly reduces sample complexity relative to the prior art by learning sub-graphs. Under the assumption of Gaussian exogenous inputs and minimum-mean squared error weight estimation, a new uncertainty bound tailored to the causal bandit problem is derived. This uncertainty bound drives an upper confidence bound based intervention selection to optimize the reward. To cope with non-stationary bandits, a sub-graph change detection mechanism is proposed, with high sample efficiency. Numerical results compare the new methodology to existing schemes and show a substantial performance improvement in both stationary and non-stationary settings. Compared to existing approaches, the proposed scheme takes 67% fewer samples to learn the causal structure and achieves an average reward gain of 85%.

With diet and nutrition apps reaching 1.4 billion users in 2022 [1], it's no surprise that health apps like MyFitnessPal, Noom, and Calorie Counter, are surging in popularity. However, one major setback [2] of nearly all nutrition applications is that users must enter food data manually, which is time-consuming and tedious. Thus, there has been an increasing demand for applications that can accurately identify food items, analyze their nutritional content, and offer dietary recommendations in real-time. This paper introduces a comprehensive system that combines advanced computer vision techniques with nutrition analysis, implemented in a versatile mobile and web application. The system is divided into three key components: 1) food detection using the YOLOv8 model, 2) nutrient analysis via the Edamam Nutrition Analysis API, and 3) personalized meal recommendations using the Edamam Meal Planning and Recipe Search APIs. Designed for both mobile and web platforms, the application ensures fast processing times with an intuitive user interface, with features such as data visualizations using Chart.js, a login system, and personalized settings for dietary preferences, allergies, and cuisine choices. Preliminary results showcase the system's effectiveness, making it a valuable tool for users to make informed dietary decisions.

Background: Breast ultrasound is prominently used in diagnosing breast tumors. At present, many automatic systems based on deep learning have been developed to help radiologists in diagnosis. However, training such systems remains challenging because they are usually data-hungry and demand amounts of labeled data, which need professional knowledge and are expensive. Methods: We adopted a triplet network and a self-supervised contrastive learning technique to learn representations from unlabeled breast ultrasound video clips. We further designed a new hard triplet loss to to learn representations that particularly discriminate positive and negative image pairs that are hard to recognize. We also constructed a pretraining dataset from breast ultrasound videos (1,360 videos from 200 patients), which includes an anchor sample dataset with 11,805 images, a positive sample dataset with 188,880 images, and a negative sample dataset dynamically generated from video clips. Further, we constructed a finetuning dataset, including 400 images from 66 patients. We transferred the pretrained network to a downstream benign/malignant classification task and compared the performance with other state-of-the-art models, including three models pretrained on ImageNet and a previous contrastive learning model retrained on our datasets. Results and conclusion: Experiments revealed that our model achieved an area under the receiver operating characteristic curve (AUC) of 0.952, which is significantly higher than the others. Further, we assessed the dependence of our pretrained model on the number of labeled data and revealed that <100 samples were required to achieve an AUC of 0.901. The proposed framework greatly reduces the demand for labeled data and holds potential for use in automatic breast ultrasound image diagnosis.

Detecting out-of-distribution (OOD) samples is a critical task for reliable machine learning. However, it becomes particularly challenging when the models are trained on long-tailed datasets, as the models often struggle to distinguish tail-class in-distribution samples from OOD samples. We examine the main challenges in this problem by identifying the trade-offs between OOD detection and in-distribution (ID) classification, faced by existing methods. We then introduce our method, called \textit{Representation Norm Amplification} (RNA), which solves this challenge by decoupling the two problems. The main idea is to use the norm of the representation as a new dimension for OOD detection, and to develop a training method that generates a noticeable discrepancy in the representation norm between ID and OOD data, while not perturbing the feature learning for ID classification. Our experiments show that RNA achieves superior performance in both OOD detection and classification compared to the state-of-the-art methods, by 1.70\% and 9.46\% in FPR95 and 2.43\% and 6.87\% in classification accuracy on CIFAR10-LT and ImageNet-LT, respectively. The code for this work is available at https://github.com/dgshin21/RNA.