Abstract
Maximizing the separation between classes constitutes a well-known inductive bias in machine learning and a pillar of many traditional algorithms. By default, deep networks are not equipped with this inductive bias and therefore many alternative solutions have been proposed through differential optimization. Current approaches tend to optimize classification and separation jointly: aligning inputs with class vectors and separating class vectors angularly. This paper proposes a simple alternative: encoding maximum separation as an inductive bias in the network by adding one fixed matrix multiplication before computing the softmax activations. The main observation behind our approach is that separation does not require optimization but can be solved in closed-form prior to training and plugged into a network. We outline a recursive approach to obtain the matrix consisting of maximally separable vectors for any number of classes, which can be added with negligible engineering effort and computational overhead. Despite its simple nature, this one matrix multiplication provides real impact. We show that our proposal directly boosts classification, long-tailed recognition, out-of-distribution detection, and open-set recognition, from CIFAR to ImageNet. We find empirically that maximum separation works best as a fixed bias; making the matrix learnable adds nothing to the performance. The closed-form implementation and code to reproduce the experiments are on github.
Keyword: expected calibration error
There is no result
Keyword: overconfident
A Comparative Study of Confidence Calibration in Deep Learning: From Computer Vision to Medical Imaging
Authors: Riqiang Gao, Thomas Li, Yucheng Tang, Zhoubing Xu, Michael Kammer, Sanja L. Antic, Kim Sandler, Fabien Moldonado, Thomas A. Lasko, Bennett Landman
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
Although deep learning prediction models have been successful in the discrimination of different classes, they can often suffer from poor calibration across challenging domains including healthcare. Moreover, the long-tail distribution poses great challenges in deep learning classification problems including clinical disease prediction. There are approaches proposed recently to calibrate deep prediction in computer vision, but there are no studies found to demonstrate how the representative models work in different challenging contexts. In this paper, we bridge the confidence calibration from computer vision to medical imaging with a comparative study of four high-impact calibration models. Our studies are conducted in different contexts (natural image classification and lung cancer risk estimation) including in balanced vs. imbalanced training sets and in computer vision vs. medical imaging. Our results support key findings: (1) We achieve new conclusions which are not studied under different learning contexts, e.g., combining two calibration models that both mitigate the overconfident prediction can lead to under-confident prediction, and simpler calibration models from the computer vision domain tend to be more generalizable to medical imaging. (2) We highlight the gap between general computer vision tasks and medical imaging prediction, e.g., calibration methods ideal for general computer vision tasks may in fact damage the calibration of medical imaging prediction. (3) We also reinforce previous conclusions in natural image classification settings. We believe that this study has merits to guide readers to choose calibration models and understand gaps between general computer vision and medical imaging domains.
Keyword: overconfidence
There is no result
Keyword: confidence
Thompson Sampling for Robust Transfer in Multi-Task Bandits
Abstract
We study the problem of online multi-task learning where the tasks are performed within similar but not necessarily identical multi-armed bandit environments. In particular, we study how a learner can improve its overall performance across multiple related tasks through robust transfer of knowledge. While an upper confidence bound (UCB)-based algorithm has recently been shown to achieve nearly-optimal performance guarantees in a setting where all tasks are solved concurrently, it remains unclear whether Thompson sampling (TS) algorithms, which have superior empirical performance in general, share similar theoretical properties. In this work, we present a TS-type algorithm for a more general online multi-task learning protocol, which extends the concurrent setting. We provide its frequentist analysis and prove that it is also nearly-optimal using a novel concentration inequality for multi-task data aggregation at random stopping times. Finally, we evaluate the algorithm on synthetic data and show that the TS-type algorithm enjoys superior empirical performance in comparison with the UCB-based algorithm and a baseline algorithm that performs TS for each individual task without transfer.
Explainability's Gain is Optimality's Loss? -- How Explanations Bias Decision-making
Authors: Charles Wan, Rodrigo Belo, Leid Zejnilović
Abstract
Decisions in organizations are about evaluating alternatives and choosing the one that would best serve organizational goals. To the extent that the evaluation of alternatives could be formulated as a predictive task with appropriate metrics, machine learning algorithms are increasingly being used to improve the efficiency of the process. Explanations help to facilitate communication between the algorithm and the human decision-maker, making it easier for the latter to interpret and make decisions on the basis of predictions by the former. Feature-based explanations' semantics of causal models, however, induce leakage from the decision-maker's prior beliefs. Our findings from a field experiment demonstrate empirically how this leads to confirmation bias and disparate impact on the decision-maker's confidence in the predictions. Such differences can lead to sub-optimal and biased decision outcomes.
A Comparative Study of Confidence Calibration in Deep Learning: From Computer Vision to Medical Imaging
Authors: Riqiang Gao, Thomas Li, Yucheng Tang, Zhoubing Xu, Michael Kammer, Sanja L. Antic, Kim Sandler, Fabien Moldonado, Thomas A. Lasko, Bennett Landman
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
Although deep learning prediction models have been successful in the discrimination of different classes, they can often suffer from poor calibration across challenging domains including healthcare. Moreover, the long-tail distribution poses great challenges in deep learning classification problems including clinical disease prediction. There are approaches proposed recently to calibrate deep prediction in computer vision, but there are no studies found to demonstrate how the representative models work in different challenging contexts. In this paper, we bridge the confidence calibration from computer vision to medical imaging with a comparative study of four high-impact calibration models. Our studies are conducted in different contexts (natural image classification and lung cancer risk estimation) including in balanced vs. imbalanced training sets and in computer vision vs. medical imaging. Our results support key findings: (1) We achieve new conclusions which are not studied under different learning contexts, e.g., combining two calibration models that both mitigate the overconfident prediction can lead to under-confident prediction, and simpler calibration models from the computer vision domain tend to be more generalizable to medical imaging. (2) We highlight the gap between general computer vision tasks and medical imaging prediction, e.g., calibration methods ideal for general computer vision tasks may in fact damage the calibration of medical imaging prediction. (3) We also reinforce previous conclusions in natural image classification settings. We believe that this study has merits to guide readers to choose calibration models and understand gaps between general computer vision and medical imaging domains.
Keyword: scaling
A IETI-DP method for discontinuous Galerkin discretizations in Isogeometric Analysis with inexact local solvers
Authors: Monica Montardini, Giancarlo Sangalli, Rainer Schneckenleitner, Stefan Takacs, Mattia Tani
Abstract
We construct solvers for an isogeometric multi-patch discretization, where the patches are coupled via a discontinuous Galerkin approach, which allows the consideration of discretizations that do not match on the interfaces. We solve the resulting linear system using a Dual-Primal IsogEometric Tearing and Interconnecting (IETI-DP) method. We are interested in solving the arising patch-local problems using iterative solvers since this allows the reduction of the memory footprint. We solve the patch-local problems approximately using the Fast Diagonalization method, which is known to be robust in the grid size and the spline degree. To obtain the tensor structure needed for the application of the Fast Diagonalization method, we introduce an orthogonal splitting of the local function spaces. We present a convergence theory that confirms that the condition number of the preconditioned system only grows poly-logarithmically with the grid size. The numerical experiments confirm this finding. Moreover, they show that the convergence of the overall solver only mildly depends on the spline degree. We observe a mild reduction of the computational times and a significant reduction of the memory requirements in comparison to standard IETI-DP solvers using sparse direct solvers for the local subproblems. Furthermore, the experiments indicate good scaling behavior on distributed memory machines.
A Parametric Class of Approximate Gradient Updates for Policy Optimization
Abstract
Approaches to policy optimization have been motivated from diverse principles, based on how the parametric model is interpreted (e.g. value versus policy representation) or how the learning objective is formulated, yet they share a common goal of maximizing expected return. To better capture the commonalities and identify key differences between policy optimization methods, we develop a unified perspective that re-expresses the underlying updates in terms of a limited choice of gradient form and scaling function. In particular, we identify a parameterized space of approximate gradient updates for policy optimization that is highly structured, yet covers both classical and recent examples, including PPO. As a result, we obtain novel yet well motivated updates that generalize existing algorithms in a way that can deliver benefits both in terms of convergence speed and final result quality. An experimental investigation demonstrates that the additional degrees of freedom provided in the parameterized family of updates can be leveraged to obtain non-trivial improvements both in synthetic domains and on popular deep RL benchmarks.
The Role of Depth, Width, and Activation Complexity in the Number of Linear Regions of Neural Networks
Authors: Alexis Goujon, Arian Etemadi, Michael Unser
Abstract
Many feedforward neural networks generate continuous and piecewise-linear (CPWL) mappings. Specifically, they partition the input domain into regions on which the mapping is an affine function. The number of these so-called linear regions offers a natural metric to characterize the expressiveness of CPWL mappings. Although the precise determination of this quantity is often out of reach, bounds have been proposed for specific architectures, including the well-known ReLU and Maxout networks. In this work, we propose a more general perspective and provide precise bounds on the maximal number of linear regions of CPWL networks based on three sources of expressiveness: depth, width, and activation complexity. Our estimates rely on the combinatorial structure of convex partitions and highlight the distinctive role of depth which, on its own, is able to exponentially increase the number of regions. We then introduce a complementary stochastic framework to estimate the average number of linear regions produced by a CPWL network architecture. Under reasonable assumptions, the expected density of linear regions along any 1D path is bounded by the product of depth, width, and a measure of activation complexity (up to a scaling factor). This yields an identical role to the three sources of expressiveness: no exponential growth with depth is observed anymore.
Bridge-Tower: Building Bridges Between Encoders in Vision-Language Representation Learning
Abstract
Vision-Language (VL) models with the Two-Tower architecture have dominated visual-language representation learning in recent years. Current VL models either use lightweight uni-modal encoders and learn to extract, align and fuse both modalities simultaneously in a cross-modal encoder, or feed the last-layer uni-modal features directly into the top cross-modal encoder, ignoring the semantic information at the different levels in the deep uni-modal encoders. Both approaches possibly restrict vision-language representation learning and limit model performance. In this paper, we introduce multiple bridge layers that build a connection between the top layers of uni-modal encoders and each layer of the cross-modal encoder. This enables comprehensive bottom-up interactions between visual and textual representations at different semantic levels, resulting in more effective cross-modal alignment and fusion. Our proposed Bridge-Tower, pre-trained with only $4$M images, achieves state-of-the-art performance on various downstream vision-language tasks. On the VQAv2 test-std set, Bridge-Tower achieves an accuracy of $78.73\%$, outperforming the previous state-of-the-art METER model by $1.09\%$ with the same pre-training data and almost no additional parameters and computational cost. Notably, when further scaling the model, Bridge-Tower achieves an accuracy of $81.15\%$, surpassing models that are pre-trained on orders-of-magnitude larger datasets. Code is available at https://github.com/microsoft/BridgeTower.
You Only Derive Once (YODO): Automatic Differentiation for Efficient Sensitivity Analysis in Bayesian Networks
Authors: Rafael Ballester-Ripoll, Manuele Leonelli
Abstract
Sensitivity analysis measures the influence of a Bayesian network's parameters on a quantity of interest defined by the network, such as the probability of a variable taking a specific value. In particular, the so-called sensitivity value measures the quantity of interest's partial derivative with respect to the network's conditional probabilities. However, finding such values in large networks with thousands of parameters can become computationally very expensive. We propose to use automatic differentiation combined with exact inference to obtain all sensitivity values in a single pass. Our method first marginalizes the whole network once using e.g. variable elimination and then backpropagates this operation to obtain the gradient with respect to all input parameters. We demonstrate our routines by ranking all parameters by importance on a Bayesian network modeling humanitarian crises and disasters, and then show the method's efficiency by scaling it to huge networks with up to 100'000 parameters. An implementation of the methods using the popular machine learning library PyTorch is freely available.
Keyword: calibration
Digital Twin Data Modelling by Randomized Orthogonal Decomposition and Deep Learning
Authors: Diana Alina Bistrian, Omer San, Ionel Michael Navon
Abstract
A digital twin is a surrogate model that has the main feature to mirror the original process behavior. Associating the dynamical process with a digital twin model of reduced complexity has the significant advantage to map the dynamics with high accuracy and reduced costs in CPU time and hardware to timescales over which that suffers significantly changes and so it is difficult to explore. This paper introduces a new framework for creating efficient digital twin models of fluid flows. We introduce a novel algorithm that combines the advantages of Krylov based dynamic mode decomposition with proper orthogonal decomposition and outperforms the selection of the most influential modes. We prove that randomized orthogonal decomposition algorithm provides several advantages over SVD empirical orthogonal decomposition methods and mitigates the projection error formulating a multiobjective optimization problem.We involve the state-of-the-art artificial intelligence Deep Learning (DL) to perform a real-time adaptive calibration of the digital twin model, with increasing fidelity. The output is a high-fidelity DIGITAL TWIN DATA MODEL of the fluid flow dynamics, with the advantage of a reduced complexity. The new modelling tools are investigated in the numerical simulation of three wave phenomena with increasing complexity. We show that the outputs are consistent with the original source data.We perform a thorough assessment of the performance of the new digital twin data models, in terms of numerical accuracy and computational efficiency, including a time simulation response feature study.
A Comparative Study of Confidence Calibration in Deep Learning: From Computer Vision to Medical Imaging
Authors: Riqiang Gao, Thomas Li, Yucheng Tang, Zhoubing Xu, Michael Kammer, Sanja L. Antic, Kim Sandler, Fabien Moldonado, Thomas A. Lasko, Bennett Landman
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
Although deep learning prediction models have been successful in the discrimination of different classes, they can often suffer from poor calibration across challenging domains including healthcare. Moreover, the long-tail distribution poses great challenges in deep learning classification problems including clinical disease prediction. There are approaches proposed recently to calibrate deep prediction in computer vision, but there are no studies found to demonstrate how the representative models work in different challenging contexts. In this paper, we bridge the confidence calibration from computer vision to medical imaging with a comparative study of four high-impact calibration models. Our studies are conducted in different contexts (natural image classification and lung cancer risk estimation) including in balanced vs. imbalanced training sets and in computer vision vs. medical imaging. Our results support key findings: (1) We achieve new conclusions which are not studied under different learning contexts, e.g., combining two calibration models that both mitigate the overconfident prediction can lead to under-confident prediction, and simpler calibration models from the computer vision domain tend to be more generalizable to medical imaging. (2) We highlight the gap between general computer vision tasks and medical imaging prediction, e.g., calibration methods ideal for general computer vision tasks may in fact damage the calibration of medical imaging prediction. (3) We also reinforce previous conclusions in natural image classification settings. We believe that this study has merits to guide readers to choose calibration models and understand gaps between general computer vision and medical imaging domains.
Edge-Aided Sensor Data Sharing in Vehicular Communication Networks
Abstract
Sensor data sharing in vehicular networks can significantly improve the range and accuracy of environmental perception for connected automated vehicles. Different concepts and schemes for dissemination and fusion of sensor data have been developed. It is common to these schemes that measurement errors of the sensors impair the perception quality and can result in road traffic accidents. Specifically, when the measurement error from the sensors (also referred as measurement noise) is unknown and time varying, the performance of the data fusion process is restricted, which represents a major challenge in the calibration of sensors. In this paper, we consider sensor data sharing and fusion in a vehicular network with both, vehicle-to-infrastructure and vehicle-to-vehicle communication. We propose a method, named Bidirectional Feedback Noise Estimation (BiFNoE), in which an edge server collects and caches sensor measurement data from vehicles. The edge estimates the noise and the targets alternately in double dynamic sliding time windows and enhances the distributed cooperative environment sensing at each vehicle with low communication costs. We evaluate the proposed algorithm and data dissemination strategy in an application scenario by simulation and show that the perception accuracy is on average improved by around 80 % with only 12 kbps uplink and 28 kbps downlink bandwidth.
Keyword: out of distribution detection
There is no result
Keyword: out-of-distribution detection
Maximum Class Separation as Inductive Bias in One Matrix
Keyword: expected calibration error
There is no result
Keyword: overconfident
A Comparative Study of Confidence Calibration in Deep Learning: From Computer Vision to Medical Imaging
Keyword: overconfidence
There is no result
Keyword: confidence
Thompson Sampling for Robust Transfer in Multi-Task Bandits
Explainability's Gain is Optimality's Loss? -- How Explanations Bias Decision-making
A Comparative Study of Confidence Calibration in Deep Learning: From Computer Vision to Medical Imaging
Keyword: scaling
A IETI-DP method for discontinuous Galerkin discretizations in Isogeometric Analysis with inexact local solvers
A Parametric Class of Approximate Gradient Updates for Policy Optimization
The Role of Depth, Width, and Activation Complexity in the Number of Linear Regions of Neural Networks
Bridge-Tower: Building Bridges Between Encoders in Vision-Language Representation Learning
You Only Derive Once (YODO): Automatic Differentiation for Efficient Sensitivity Analysis in Bayesian Networks
Keyword: calibration
Digital Twin Data Modelling by Randomized Orthogonal Decomposition and Deep Learning
A Comparative Study of Confidence Calibration in Deep Learning: From Computer Vision to Medical Imaging
Edge-Aided Sensor Data Sharing in Vehicular Communication Networks