Quantifying Local Specialization in Deep Neural Networks
University of California Berkeley（加州大学伯克利分校）；Boston University（波士顿大学）；Center for Human-Compatible AI (CHAI)（人类兼容人工智能中心）；MIT Computer Science and Artificial Intelligence Laboratory (CSAIL)（麻省理工学院计算机科学与人工智能实验室）
Shlomi Hod, Daniel Filan, Stephen Casper, Andrew Critch, Stuart Russell
A neural network is locally specialized to the extent that parts of its computational graph (i.e. structure) can be abstractly represented as performing some comprehensible sub-task relevant to the overall task (i.e. functionality). Are modern deep neural networks locally specialized? How can this be quantified? In this paper, we consider the problem of taking a neural network whose neurons are partitioned into clusters, and quantifying how functionally specialized the clusters are. We propose two proxies for this: importance, which reflects how crucial sets of neurons are to network performance; and coherence, which reflects how consistently their neurons associate with features of the inputs. To measure these proxies, we develop a set of statistical methods based on techniques conventionally used to interpret individual neurons. We apply the proxies to partitionings generated by spectrally clustering a graph representation of the network's neurons with edges determined either by network weights or correlations of activations. We show that these partitionings, even ones based only on weights (i.e. strictly from non-runtime analysis), reveal groups of neurons that are important and coherent. These results suggest that graph-based partitioning can reveal local specialization and that statistical methods can be used to automatedly screen for sets of neurons that can be understood abstractly.
University of Cambridge（剑桥大学）；University of California, Los Angeles（加利福尼亚大学洛杉矶分校）；Alan Turing Institute（艾伦·图灵研究所）
Despite their success with unstructured data, deep neural networks are not yet a panacea for structured tabular data. In the tabular domain, their efficiency crucially relies on various forms of regularization to prevent overfitting and provide strong generalization performance. Existing regularization techniques include broad modelling decisions such as choice of architecture, loss functions, and optimization methods. In this work, we introduce Tabular Neural Gradient Orthogonalization and Specialization (TANGOS), a novel framework for regularization in the tabular setting built on latent uni…
SD-MoE: Spectral Decomposition for Effective Expert Specialization
College of Computer Science and Artificial Intelligence, Fudan University, Shanghai, China（复旦大学计算机科学与人工智能学院）；University of Bath, Bath, United Kingdom（巴斯大学）；Oxford Suzhou Centre for Advanced Research, Suzhou, China（牛津苏泽研究中心）；Department of Electrical Engineering and Computer Science, University of Michigan（密歇根大学电气工程与计算机科学系）；Shanghai Innovation Institute, Shanghai, China（上海创新研究院）；Department of Computer Science, University of Colorado Boulder, Colorado, USA（科罗拉多大学博尔德分校计算机科学系）；Research Institute of Tsinghua University in Shenzhen, Shenzhen, China（清华大学深圳研究院）；Greater Bay Area National Center of Technology Innovation, Research Institute of Tsinghua University in Shenzhen, Shenzhen, China（粤港澳大湾区国家技术创新中心，清华大学深圳研究院）；School of Microelectronics, Fudan University, Shanghai, China（复旦大学微电子学院）
Texas State University（得克萨斯州立大学）；Illinois Institute of Technology（伊利诺伊理工学院）
Algorithmic bias is of increasing concern, both to the research community, and society at large. Bias in AI is more abstract and unintuitive than traditional forms of discrimination and can be more difficult to detect and mitigate. A clear gap exists in the current literature on evaluating the relative bias in the performance of multi-class classifiers. In this work, we propose two simple yet effective metrics, Combined Error Variance (CEV) and Symmetric Distance Error (SDE), to quantitatively evaluate the class-wise bias of two models in comparison to one another. By evaluating the performanc…
Post-hoc analysis is a popular category in eXplainable artificial intelligence (XAI) study. In particular, methods that generate heatmaps have been used to explain the deep neural network (DNN), a black-box model. Heatmaps can be appealing due to the intuitive and visual ways to understand them but assessing their qualities might not be straightforward. Different ways to assess heatmaps' quality have their own merits and shortcomings. This paper introduces a synthetic dataset that can be generated adhoc along with the ground-truth heatmaps for more objective quantitative assessment. Each sampl…
Northeastern University（东北大学）；risQ Inc.（risQ公司）；Bay Area Environmental Research Institute（湾区环境研究所）；NASA Ames Research Center（美国国家航空航天局艾姆斯研究中心）；NASA Advanced Supercomputing Division（美国国家航空航天局高级超级计算部）
Deep Learning (DL) methods have been transforming computer vision with innovative adaptations to other domains including climate change. For DL to pervade Science and Engineering (S&E) applications where risk management is a core component, well-characterized uncertainty estimates must accompany predictions. However, S&E observations and model-simulations often follow heavily skewed distributions and are not well modeled with DL approaches, since they usually optimize a Gaussian, or Euclidean, likelihood loss. Recent developments in Bayesian Deep Learning (BDL), which attempts to capture uncer…
Quantifying Causes of Arctic Amplification via Deep Learning based Time-series Causal Inference
University of Maryland, Baltimore County（马里兰大学巴尔的摩郡分校）；University of Colorado Boulder（科罗拉多大学博尔德分校）；Geophysical Fluid Dynamics Laboratory, NOAA（美国国家海洋和大气管理局地球物理流体动力学实验室）；NSF HDR Institute for Harnessing Data and Model Revolution in the Polar Regions (iHARP)（美国国家科学基金会HDR极地数据与模型变革应用研究所（iHARP））
University of Waterloo（滑铁卢大学）；Waterloo Artificial Intelligence Institute（滑铁卢人工智能研究所）；DarwinAI Corp.（DarwinAI公司）
The quantitative analysis of information structure through a deep neural network (DNN) can unveil new insights into the theoretical performance of DNN architectures. Two very promising avenues of research towards quantitative information structure analysis are: 1) layer similarity (LS) strategies focused on the inter-layer feature similarity, and 2) intrinsic dimensionality (ID) strategies focused on layer-wise data dimensionality using pairwise information. Inspired by both LS and ID strategies for quantitative information structure analysis, we introduce two novel complimentary methods for i…
In an effort to understand the meaning of the intermediate representations captured by deep networks, recent papers have tried to associate specific semantic concepts to individual neural network filter responses, where interesting correlations are often found, largely by focusing on extremal filter responses. In this paper, we show that this approach can favor easy-to-interpret cases that are not necessarily representative of the average behavior of a representation. A more realistic but harder-to-study hypothesis is that semantic representations are distributed, and thus filters must be stud…
TU Berlin（柏林工业大学）；MPII（马克斯·普朗克信息学研究所）；Korea University（高丽大学）；Google Research（谷歌研究院）；RIKEN AIP（理化学研究所人工智能项目中心）；TU Kaiserslautern（凯泽斯劳滕工业大学）
Explainable AI (XAI) aims to provide interpretations for predictions made by learning machines, such as deep neural networks, in order to make the machines more transparent for the user and furthermore trustworthy also for applications in e.g. safety-critical areas. So far, however, no methods for quantifying uncertainties of explanations have been conceived, which is problematic in domains where a high confidence in explanations is a prerequisite. We therefore contribute by proposing a new framework that allows to convert any arbitrary explanation method for neural networks into an explanatio…
Improving accuracy and uncertainty quantification of deep learning based quantitative MRI using Monte Carlo dropout
Technical University of Munich（慕尼黑工业大学）；University of Oxford（牛津大学）；Massachusetts General Hospital（麻省总医院）；Harvard Medical School（哈佛医学院）；Tianjin University（天津大学）
针对深度学习定量 MRI 的精度与不确定性问题，提出训练和推理均使用 Monte Carlo dropout 并平均多次预测的方法，在小数据集下显著提升 FA、MD 图精度，还能生成辅助诊断的置信图。
We consider a deep ReLU / Leaky ReLU student network trained from the output of a fixed teacher network of the same depth, with Stochastic Gradient Descent (SGD). The student network is \emph{over-realized}: at each layer $l$, the number $n_l$ of student nodes is more than that ($m_l$) of teacher. Under mild conditions on dataset and teacher network, we prove that when the gradient is small at every data sample, each teacher node is \emph{specialized} by at least one student node \emph{at the lowest layer}. For two-layer network, such specialization can be achieved by training on any dataset o…
University of Washington（华盛顿大学）；Shanghai Jiao Tong University（上海交通大学）
Specialized Deep Learning (DL) acceleration stacks, designed for a specific set of frameworks, model architectures, operators, and data types, offer the allure of high performance while sacrificing flexibility. Changes in algorithms, models, operators, or numerical systems threaten the viability of specialized hardware accelerators. We propose VTA, a programmable deep learning architecture template designed to be extensible in the face of evolving workloads. VTA achieves this flexibility via a parametrizable architecture, two-level ISA, and a JIT compiler. The two-level ISA is based on (1) a t…
Lab126, Amazon（亚马逊Lab126）；Open University of Israel（以色列开放大学）
We propose a method for specializing deep detectors and trackers to restricted settings. Our approach is designed with the following goals in mind: (a) Improving accuracy in restricted domains; (b) preventing overfitting to new domains and forgetting of generalized capabilities; (c) aggressive model compression and acceleration. To this end, we propose a novel loss that balances compression and acceleration of a deep learning model vs. loss of generalization capabilities. We apply our method to the existing tracker and detector models. We report detection results on the VIRAT and CAVIAR data s…
We propose two bounded comparison metrics that may be implemented to arbitrary dimensions in regression tasks. One quantifies the structure of uncertainty and the other quantifies the distribution of uncertainty. The structure metric assesses the similarity in shape and location of uncertainty with the true error, while the distribution metric quantifies the supported magnitudes between the two. We apply these metrics to Gaussian Processes (GPs), Ensemble Deep Neural Nets (DNNs), and Ensemble Deep Neural Operators (DNOs) on high-dimensional and nonlinear test cases. We find that comparing a mo…
We prove limitations on what neural networks trained by noisy gradient descent (GD) can efficiently learn. Our results apply whenever GD training is equivariant, which holds for many standard architectures and initializations. As applications, (i) we characterize the functions that fully-connected networks can weak-learn on the binary hypercube and unit sphere, demonstrating that depth-2 is as powerful as any other depth for this task; (ii) we extend the merged-staircase necessity result for learning with latent low-dimensional structure [ABM22] to beyond the mean-field regime. Under cryptogra…
College of Aerospace Science and Engineering National University of Defense Technology（国防科技大学航空航天学院）；Defense Innovation Institute Chinese Academy of Military Science（中国人民解放军军事科学院国防创新研究院）
For the temperature field reconstruction (TFR), a complex image-to-image regression problem, the convolutional neural network (CNN) is a powerful surrogate model due to the convolutional layer's good image feature extraction ability. However, a lot of labeled data is needed to train CNN, and the common CNN can not quantify the aleatoric uncertainty caused by data noise. In actual engineering, the noiseless and labeled training data is hardly obtained for the TFR. To solve these two problems, this paper proposes a deep Monte Carlo quantile regression (Deep MC-QR) method for reconstructing the t…
Deep learning models achieve high predictive accuracy across a broad spectrum of tasks, but rigorously quantifying their predictive uncertainty remains challenging. Usable estimates of predictive uncertainty should (1) cover the true prediction targets with high probability, and (2) discriminate between high- and low-confidence prediction instances. Existing methods for uncertainty quantification are based predominantly on Bayesian neural networks; these may fall short of (1) and (2) -- i.e., Bayesian credible intervals do not guarantee frequentist coverage, and approximate posterior inference…
UC San Diego（加州大学圣地亚哥分校）；University of Washington（华盛顿大学）；Northeastern University（东北大学）
Deep learning is gaining increasing popularity for spatiotemporal forecasting. However, prior works have mostly focused on point estimates without quantifying the uncertainty of the predictions. In high stakes domains, being able to generate probabilistic forecasts with confidence intervals is critical to risk assessment and decision making. Hence, a systematic study of uncertainty quantification (UQ) methods for spatiotemporal forecasting is missing in the community. In this paper, we describe two types of spatiotemporal forecasting problems: regular grid-based and graph-based. Then we analyz…
University of California, Davis（加州大学戴维斯分校）；New York University（纽约大学）；University of California, Berkeley（加州大学伯克利分校）
With the widespread success of deep neural networks in science and technology, it is becoming increasingly important to quantify the uncertainty of the predictions produced by deep learning. In this paper, we introduce a new method that attaches an explicit uncertainty statement to the probabilities of classification using deep neural networks. Precisely, we view that the classification probabilities are sampled from an unknown distribution, and we propose to learn this distribution through the Dirichlet mixture that is flexible enough for approximating any continuous distribution on the simpl…
Deep neural networks are increasingly being used for the analysis of medical images. However, most works neglect the uncertainty in the model's prediction. We propose an uncertainty-aware deep kernel learning model which permits the estimation of the uncertainty in the prediction by a pipeline of a Convolutional Neural Network and a sparse Gaussian Process. Furthermore, we adapt different pre-training methods to investigate their impacts on the proposed model. We apply our approach to Bone Age Prediction and Lesion Localization. In most cases, the proposed model shows better performance compar…
Recent advances in reconstruction methods for inverse problems leverage powerful data-driven models, e.g., deep neural networks. These techniques have demonstrated state-of-the-art performances for several imaging tasks, but they often do not provide uncertainty on the obtained reconstruction. In this work, we develop a scalable, data-driven, knowledge-aided computational framework to quantify the model uncertainty via Bayesian neural networks. The approach builds on, and extends deep gradient descent, a recently developed greedy iterative training scheme, and recasts it within a probabilistic…
Government Technology Agency of Singapore（新加坡政府科技局）；Temple University（天普大学）；Senseable City Lab（可感知城市实验室）
Urban canopy cover is important to mitigate the impact of climate change. Yet, existing quantification of urban greenery is either manual and not scalable, or use traditional computer vision methods that are inaccurate. We train deep convolutional neural networks (DCNNs) on datasets used for self-driving cars to estimate urban greenery instead, and find that our semantic segmentation and direct end-to-end estimation method are more accurate and scalable, reducing mean absolute error of estimating the Green View Index (GVI) metric from 10.1% to 4.67%. With the revised DCNN methods, the Treepedi…
Queen Mary University of London（伦敦玛丽女王大学）；Imperial College London（帝国理工学院）；Facebook（脸书）
We show that the classification performance of graph convolutional networks (GCNs) is related to the alignment between features, graph, and ground truth, which we quantify using a subspace alignment measure (SAM) corresponding to the Frobenius norm of the matrix of pairwise chordal distances between three subspaces associated with features, graph, and ground truth. The proposed measure is based on the principal angles between subspaces and has both spectral and geometrical interpretations. We showcase the relationship between the SAM and the classification performance through the study of limi…
Mila, Quebec Artificial Intelligence Institute（米拉-魁北克人工智能研究所）；Université de Montréal（蒙特利尔大学）
Most deep learning models in chest X-ray prediction utilize the posteroanterior (PA) view due to the lack of other views available. PadChest is a large-scale chest X-ray dataset that has almost 200 labels and multiple views available. In this work, we use PadChest to explore multiple approaches to merging the PA and lateral views for predicting the radiological labels associated with the X-ray image. We find that different methods of merging the model utilize the lateral view differently. We also find that including the lateral view increases performance for 32 labels in the dataset, while bei…
Friedrich-Alexander University Erlangen (FAU)（埃尔朗根-纽伦堡弗里德里希-亚历山大大学（FAU））；University Hospital Erlangen（埃尔朗根大学医院）
Deep neural networks typically outperform more traditional machine learning models in their ability to classify complex data, and yet is not clear how the individual hidden layers of a deep network contribute to the overall classification performance. We thus introduce a Generalized Discrimination Value (GDV) that measures, in a non-invasive manner, how well different data classes separate in each given network layer. The GDV can be used for the automatic tuning of hyper-parameters, such as the width profile and the total depth of a network. Moreover, the layer-dependent GDV(L) provides new in…
Understanding the underlying mechanisms that enable the empirical successes of deep neural networks is essential for further improving their performance and explaining such networks. Towards this goal, a specific question is how to explain the "surprising" behavior of the same over-parametrized deep neural networks that can generalize well on real datasets and at the same time "memorize" training samples when the labels are randomized. In this paper, we demonstrate that deep ReLU networks generalize from training samples to new points via piece-wise linear interpolation. We provide a quantifie…
In this paper, we define the term 'DigitalExposome' as a conceptual framework that takes us closer towards understanding the relationship between environment, personal characteristics, behaviour and wellbeing using multimodel mobile sensing technology. Specifically, we simultaneously collected (for the first time) multi-sensor data including urban environmental factors (e.g. air pollution including: PM1, PM2.5, PM10, Oxidised, Reduced, NH3 and Noise, People Count in the vicinity), body reaction (physiological reactions including: EDA, HR, HRV, Body Temperature, BVP and movement) and individual…
Samsung AI Center, Cambridge（三星人工智能中心剑桥）；University of Oxford（牛津大学）
In recent years, advances in deep learning have resulted in unprecedented leaps in diverse tasks spanning from speech and object recognition to context awareness and health monitoring. As a result, an increasing number of AI-enabled applications are being developed targeting ubiquitous and mobile devices. While deep neural networks (DNNs) are getting bigger and more complex, they also impose a heavy computational and energy burden on the host devices, which has led to the integration of various specialized processors in commodity devices. Given the broad range of competing DNN architectures an…
CONNECT-AI Research Center, Yonsei University College of Medicine（延世大学医学院CONNECT-AI研究中心）；Gordon Center for Medical Imaging, Massachusetts General Hospital（麻省总医院戈登医学影像中心）；Harvard Medical School（哈佛医学院）
Medical imaging, including MRI, CT, and Ultrasound, plays a vital role in clinical decisions. Accurate segmentation is essential to measure the structure of interest from the image. However, manual segmentation is highly operator-dependent, which leads to high inter and intra-variability of quantitative measurements. In this paper, we explore the feasibility that Bayesian predictive distribution parameterized by deep neural networks can capture the clinicians' inter-intra variability. By exploring and analyzing recently emerged approximate inference schemes, we evaluate whether approximate Bay…
From Heuristics to Data: Quantifying Site Planning Layout Indicators with Deep Learning and Multi-Modal Data
National University of Singapore (NUS)（新加坡国立大学）；The Cambridge Centre for Advanced Research and Education in Singapore (CARES)（新加坡剑桥高级研究与教育中心）；China Southwest Architectural Design and Research Institute Co., Ltd. (CSWADI)（中国西南建筑规划设计研究院有限公司）
SLAC National Accelerator Laboratory（SLAC国家加速器实验室）；University of Oxford（牛津大学）
Virtual Diagnostic (VD) is a deep learning tool that can be used to predict a diagnostic output. VDs are especially useful in systems where measuring the output is invasive, limited, costly or runs the risk of damaging the output. Given a prediction, it is necessary to relay how reliable that prediction is. This is known as 'uncertainty quantification' of a prediction. In this paper, we use ensemble methods and quantile regression neural networks to explore different ways of creating and analyzing prediction's uncertainty on experimental data from the Linac Coherent Light Source at SLAC. We ai…
Academy of Mathematics and Systems Science, Chinese Academy of Sciences（中国科学院数学与系统科学研究院）；University of Chinese Academy of Sciences（中国科学院大学）；Brown University（布朗大学）
The accuracy of deep learning, i.e., deep neural networks, can be characterized by dividing the total error into three main types: approximation error, optimization error, and generalization error. Whereas there are some satisfactory answers to the problems of approximation and optimization, much less is known about the theory of generalization. Most existing theoretical works for generalization fail to explain the performance of neural networks in practice. To derive a meaningful bound, we study the generalization error of neural networks for classification problems in terms of data distribut…
SAP（SAP公司）；Salzburg University of Applied Sciences（萨尔茨堡应用科学大学）；University of Heidelberg（海德堡大学）；University of Waterloo（滑铁卢大学）
Differential privacy allows bounding the influence that training data records have on a machine learning model. To use differential privacy in machine learning, data scientists must choose privacy parameters $(ε,δ)$. Choosing meaningful privacy parameters is key, since models trained with weak privacy parameters might result in excessive privacy leakage, while strong privacy parameters might overly degrade model utility. However, privacy parameter values are difficult to choose for two main reasons. First, the theoretical upper bound on privacy loss $(ε,δ)$ might be loose, depending on the cho…
Julius-Maximilians-Universität Würzburg（维尔茨堡大学）；Max Planck Institute for the Physics of Complex Systems（马克斯·普朗克复杂系统物理研究所）；Nordita KTH Royal Institute of Technology and Stockholm University（北欧理论物理研究所、瑞典皇家理工学院和斯德哥尔摩大学）
We investigate the analogy between the renormalization group (RG) and deep neural networks, wherein subsequent layers of neurons are analogous to successive steps along the RG. In particular, we quantify the flow of information by explicitly computing the relative entropy or Kullback-Leibler divergence in both the one- and two-dimensional Ising models under decimation RG, as well as in a feedforward neural network as a function of depth. We observe qualitatively identical behavior characterized by the monotonic increase to a parameter-dependent asymptotic value. On the quantum field theory sid…
University of Manchester（曼彻斯特大学）；Alan Turing Institute（艾伦·图灵研究所）
In this work we use variational inference to quantify the degree of uncertainty in deep learning model predictions of radio galaxy classification. We show that the level of model posterior variance for individual test samples is correlated with human uncertainty when labelling radio galaxies. We explore the model performance and uncertainty calibration for different weight priors and suggest that a sparse prior produces more well-calibrated uncertainty estimates. Using the posterior distributions for individual weights, we demonstrate that we can prune 30% of the fully-connected layer weights …
EvidentialRAG: Quantifying and Mitigating Information Conflict in Multi-Source Retrieval-Augmented Generation via Evidential Deep Learning
School of Computing, Wichita State University（威奇托州立大学计算学院）；Department of Computer Science, American International University-Bangladesh（美国国际孟加拉大学计算机科学系）
Argonne National Laboratory（阿贡国家实验室）；Université Paris-Saclay（巴黎萨克雷大学）
Classical problems in computational physics such as data-driven forecasting and signal reconstruction from sparse sensors have recently seen an explosion in deep neural network (DNN) based algorithmic approaches. However, most DNN models do not provide uncertainty estimates, which are crucial for establishing the trustworthiness of these techniques in downstream decision making tasks and scenarios. In recent years, ensemble-based methods have achieved significant success for the uncertainty quantification in DNNs on a number of benchmark problems. However, their performance on real-world appli…
It is widely believed that deep neural networks contain layer specialization, wherein neural networks extract hierarchical features representing edges and patterns in shallow layers and complete objects in deeper layers. Unlike common feed-forward models that have distinct filters at each layer, recurrent networks reuse the same parameters at various depths. In this work, we observe that recurrent models exhibit the same hierarchical behaviors and the same performance benefits with depth as feed-forward networks despite reusing the same filters at every recurrence. By training models of variou…
University of California, Berkeley（加利福尼亚大学伯克利分校）；Boston University（波士顿大学）
The learned weights of a neural network are often considered devoid of scrutable internal structure. To discern structure in these weights, we introduce a measurable notion of modularity for multi-layer perceptrons (MLPs), and investigate the modular structure of MLPs trained on datasets of small images. Our notion of modularity comes from the graph clustering literature: a "module" is a set of neurons with strong internal connectivity but weak external connectivity. We find that training and weight pruning produces MLPs that are more modular than randomly initialized ones, and often significa…
