Learning to predict with supporting evidence

Abstract

The impact of machine learning models on healthcare will depend on the degree of trust that healthcare professionals place in the predictions made by these models. In this paper, we present a method to provide individuals with clinical expertise with domain-relevant evidence about why a prediction should be trusted. We first design a probabilistic model that relates meaningful latent concepts to prediction targets and observed data. Inference of latent variables in this model corresponds to both making a prediction and providing supporting evidence for that prediction. We present a two-step process to efficiently approximate inference: (i) estimating model parameters using variational learning, and (ii) approximating maximum a posteriori estimation of latent variables in the model using a neural network, trained with an objective derived from the probabilistic model. We demonstrate the method on the task of predicting mortality risk for patients with cardiovascular disease. Specifically, using electrocardiogram and tabular data as input, we show that our approach provides appropriate domain-relevant supporting evidence for accurate predictions.

Meta Learning to Bridge Vision and Language Models for Multimodal Few-Shot Learning

“…Table 1: Comparison with the Frozen (Tsimpoukelli et al, 2021) baselines on Real-Name and Open-Ended miniImageNet 2-and 5-way setting; expressed in accuracy(%). ANIL (Raghu et al, 2019) is used as an upper bound, since it is a discriminative approach as opposed to our open-ended generative one. Our episodically trained models are outperforming the Frozen baselines, both with and without domain-shift.…”

“…We believe that the open-ended approach is promising due to its flexibility in reasoning about visual concepts in an unconstrained manner, instead of relying on a pre-defined closed set of concepts. However, due to the magnitudes larger hypothesis space in the open-ended text generation (the full vocabulary of the language model), compared to the finite set of possible answers of conventional classifiers, such as ANIL (Raghu et al, 2019), it is not possible to perform a fair comparison to them. Therefore, we use their results as upper bound to our approach.…”

“…After the meta-training is finished, for a new given task, the meta-trained model is used for inference by further adapting the meta-learned model with the support set, and measuring the performance on unseen query samples. in unimodal settings, particularly for few-shot image classification (Vinyals et al, 2016;Ravi & Larochelle, 2017;Finn et al, 2017;Raghu et al, 2019;Ye et al, 2020), meta-learning remains almost unexplored for multimodal few-shot settings. Similar to unimodal settings, empowering a multimodal few-shot learner with the ability to accrue knowledge across tasks, would assist in building internal multimodal representations broadly suitable for many tasks.…”

LifeLearner: Hardware-Aware Meta Continual Learning System for Embedded Computing Platforms

“…Another learning approach, called meta-learning, is proposed to make DNN more sample-efficient [15,29,53,99], requiring only a few samples to adapt/learn new data distributions from a correlated data stream [1,68]. However, existing meta-learning methods often neglect the forgetting problem of the already learned classes as it primarily aims at fast adaptation towards new tasks only [9,19,22,24,30,79,86,93].…”

The exact asymptotic form of Bayesian generalization error in latent Dirichlet allocation

“…For instance, following [28], when we predict the knee arthritis grades of patients by using x-ray images and a CBM, we set the concepts as clinical findings corrected by medical doctors, and thereby understand how clinical findings affect the predicted grades based on the correlation, by observing the learned weights in the last connection. Concept-based interpretation is used in knowledge discovery for chess [39], video representation [44], medical imaging [25], clinical risk prediction [45], computer aided diagnosis [27], and other healthcare domain problems [10]. CBM is a significant foundation for these applications, and advanced methods [49,44,27] have been proposed based on CBM.…”