ICPR2020 Paper Browser

Paper download is intended for registered attendees only, and is subjected to the IEEE Copyright Policy. Any other use is strongly forbidden.

Rethinking Experience Replay: A Bag of Tricks for Continual Learning

Pietro Buzzega, Matteo Boschini, Angelo Porrello, Simone Calderara

Auto-TLDR; Experience Replay for Continual Learning: A Practical Approach

Abstract Slides Poster

In Continual Learning, a Neural Network is trained on a stream of data whose distribution shifts over time. Under these assumptions, it is especially challenging to improve on classes appearing later in the stream while remaining accurate on previous ones. This is due to the infamous problem of catastrophic forgetting, which causes a quick performance degradation when the classifier focuses on learning new categories. Recent literature proposed various approaches to tackle this issue, often resorting to very sophisticated techniques. In this work, we show that naive rehearsal can be patched to achieve similar performance. We point out some shortcomings that restrain Experience Replay (ER) and propose five tricks to mitigate them. Experiments show that ER, thus enhanced, displays an accuracy gain of 51.2 and 26.9 percentage points on the CIFAR-10 and CIFAR-100 datasets respectively (memory buffer size 1000). As a result, it surpasses current state-of-the-art rehearsal-based methods.

Similar papers

Class-Incremental Learning with Pre-Allocated Fixed Classifiers

Federico Pernici, Matteo Bruni, Claudio Baecchi, Francesco Turchini, Alberto Del Bimbo

Auto-TLDR; Class-Incremental Learning with Pre-allocated Output Nodes for Fixed Classifier

Abstract Slides Poster Similar

In class-incremental learning, a learning agent faces a stream of data with the goal of learning new classes while not forgetting previous ones. Neural networks are known to suffer under this setting, as they forget previously acquired knowledge. To address this problem, effective methods exploit past data stored in an episodic memory while expanding the final classifier nodes to accommodate the new classes. In this work, we substitute the expanding classifier with a novel fixed classifier in which a number of pre-allocated output nodes are subject to the classification loss right from the beginning of the learning phase. Contrarily to the standard expanding classifier, this allows: (a) the output nodes of future unseen classes to firstly see negative samples since the beginning of learning together with the positive samples that incrementally arrive; (b) to learn features that do not change their geometric configuration as novel classes are incorporated in the learning model. Experiments with public datasets show that the proposed approach is as effective as the expanding classifier while exhibiting intriguing properties of internal feature representation that are otherwise not-existent. Our ablation study on pre-allocating a large number of classes further validates the approach.

Semi-Supervised Class Incremental Learning

Alexis Lechat, Stéphane Herbin, Frederic Jurie

Auto-TLDR; incremental class learning with non-annotated batches

Abstract Slides Poster Similar

This paper makes a contribution to the problem of incremental class learning, the principle of which is to sequentially introduce batches of samples annotated with new classes during the learning phase. The main objective is to reduce the drop in classification performance on old classes, a phenomenon commonly called catastrophic forgetting. We propose in this paper a new method which exploits the availability of a large quantity of non-annotated images in addition to the annotated batches. These images are used to regularize the classifier and give the feature space a more stable structure. We demonstrate on several image data sets that our approach is able to improve the global performance of classifiers learned using an incremental learning protocol, even with annotated batches of small size.

Selecting Useful Knowledge from Previous Tasks for Future Learning in a Single Network

Feifei Shi, Peng Wang, Zhongchao Shi, Yong Rui

Auto-TLDR; Continual Learning with Gradient-based Threshold Threshold

Abstract Slides Poster Similar

Continual learning is able to learn new tasks incrementally while avoiding catastrophic forgetting. Recent work has shown that packing multiple tasks into a single network incrementally by iterative pruning and re-training network is a promising method. We build upon this idea and propose an improved version of PackNet, specifically, we propose a novel gradient-based threshold method to reuse the knowledge of the previous tasks selectively when learning new tasks. Our experiments on a variety of classification tasks and different network architectures demonstrate that our method obtains competitive results when compared to PackNet.

RSAC: Regularized Subspace Approximation Classifier for Lightweight Continuous Learning

Chih-Hsing Ho, Shang-Ho Tsai

Auto-TLDR; Regularized Subspace Approximation Classifier for Lightweight Continuous Learning

Abstract Slides Poster Similar

Continuous learning seeks to perform the learning on the data that arrives from time to time. While prior works have demonstrated several possible solutions, these approaches require excessive training time as well as memory usage. This is impractical for applications where time and storage are constrained, such as edge computing. In this work, a novel training algorithm, regularized subspace approximation classifier (RSAC), is proposed to achieve lightweight continuous learning. RSAC contains a feature reduction module and classifier module with regularization. Extensive experiments show that RSAC is more efficient than prior continuous learning works and outperforms these works on various experimental settings.

Class-Incremental Learning with Topological Schemas of Memory Spaces

Xinyuan Chang, Xiaoyu Tao, Xiaopeng Hong, Xing Wei, Wei Ke, Yihong Gong

Auto-TLDR; Class-incremental Learning with Topological Schematic Model

Abstract Slides Poster Similar

Class-incremental learning (CIL) aims to incrementally learn a unified classifier for new classes emerging, which suffers from the catastrophic forgetting problem. To alleviate forgetting and improve the recognition performance, we propose a novel CIL framework, named the topological schemas model (TSM). TSM consists of a Gaussian mixture model arranged on 2D grids (2D-GMM) as the memory of the learned knowledge. To train the 2D-GMM model, we develop a novel competitive expectation-maximization (CEM) method, which contains a global topology embedding step and a local expectation-maximization finetuning step. Meanwhile, we choose the image samples of old classes that have the maximum posterior probability with respect to each Gaussian distribution as the episodic points. When finetuning for new classes, we propose the memory preservation loss (MPL) term to ensure episodic points still have maximum probabilities with respect to the corresponding Gaussian distribution. MPL preserves the distribution of 2D-GMM for old knowledge during incremental learning and alleviates catastrophic forgetting. Comprehensive experimental evaluations on two popular CIL benchmarks CIFAR100 and subImageNet demonstrate the superiority of our TSM.

ARCADe: A Rapid Continual Anomaly Detector

Ahmed Frikha, Denis Krompass, Volker Tresp

Auto-TLDR; ARCADe: A Meta-Learning Approach for Continuous Anomaly Detection

Abstract Slides Poster Similar

Although continual learning and anomaly detection have separately been well-studied in previous works, their intersection remains rather unexplored. The present work addresses a learning scenario where a model has to incrementally learn a sequence of anomaly detection tasks, i.e. tasks from which only examples from the normal (majority) class are available for training. We define this novel learning problem of continual anomaly detection (CAD) and formulate it as a meta-learning problem. Moreover, we propose \emph{A Rapid Continual Anomaly Detector (ARCADe)}, an approach to train neural networks to be robust against the major challenges of this new learning problem, namely catastrophic forgetting and overfitting to the majority class. The results of our experiments on three datasets show that, in the CAD problem setting, ARCADe substantially outperforms baselines from the continual learning and anomaly detection literature. Finally, we provide deeper insights into the learning strategy yielded by the proposed meta-learning algorithm.

Naturally Constrained Online Expectation Maximization

Daniela Pamplona, Antoine Manzanera

Auto-TLDR; Constrained Online Expectation-Maximization for Probabilistic Principal Components Analysis

Abstract Slides Poster Similar

With the rise of big data sets, learning algorithms must be adapted to piece-wise mechanisms in order to tackle time and memory costs of large scale calculations. Furthermore, for most learning embedded systems the input data are fed in a sequential and contingent manner: one by one, and possibly class by class. Thus, learning algorithms should not only run online but cope with time-varying, non-independent, and non-balanced training data for the system's entire life. Online Expectation-Maximization is a well-known algorithm for learning probabilistic models in real-time, due to its simplicity and convergence properties. However, these properties are only valid in the case of large, independent and identically distributed (iid) samples. In this paper, we propose to constraint the online Expectation-Maximization on the Fisher distance between the parameters. After the presentation of the algorithm, we make a thorough study of its use in Probabilistic Principal Components Analysis. First, we derive the update rules, then we analyse the effect of the constraint on major problems of online and sequential learning: convergence, forgetting and interference. Furthermore we use several algorithmic protocols: iid {\em vs} sequential data, and constraint parameters updated step-wise {\em vs} class-wise. Our results show that this constraint increases the convergence rate of online Expectation-Maximization, decreases forgetting and slightly introduces transfer learning.

Dual-Memory Model for Incremental Learning: The Handwriting Recognition Use Case

Mélanie Piot, Bérangère Bourdoulous, Aurelia Deshayes, Lionel Prevost

Auto-TLDR; A dual memory model for handwriting recognition

Abstract Poster Similar

In this paper, we propose a dual memory model inspired by neural science. Short-term memory processes the data stream before integrating them into long-term memory, which generalizes. The use case is learning the ability to recognize handwriting. This begins with the learning of prototypical letters . It continues throughout life and gives the individual the ability to recognize increasingly varied handwriting. This second task is achieved by incrementally training our dual-memory model. We used a convolution network for encoding and random forests as the memory model. Indeed, the latter have the advantage of being easily enhanced to integrate new data and new classes. Performances on the MNIST database are very encouraging since they exceed 95\% and the complexity of the model remains reasonable.

Learning with Delayed Feedback

Pranavan Theivendiram, Terence Sim

Auto-TLDR; Unsupervised Machine Learning with Delayed Feedback

Abstract Slides Poster Similar

We propose a novel supervised machine learning strategy, inspired by human learning, that enables an Agent to learn continually over its lifetime. A natural consequence is that the Agent must be able to handle an input whose label is delayed until a later time, or may not arrive at all. Our Agent learns in two steps: a short Seeding phase, in which the Agent's model is initialized with labelled inputs, and an indefinitely long Growing phase, in which the Agent refines and assesses its model if the label is given for an input, but stores the input in a finite-length queue if the label is missing. Queued items are matched against future input-label pairs that arrive, and the model is then updated. Our strategy also allows for the delayed feedback to take a different form. For example, in an image captioning task, the feedback could be a semantic segmentation rather than a textual caption. We show with many experiments that our strategy enables an Agent to learn flexibly and efficiently.

A Close Look at Deep Learning with Small Data

Lorenzo Brigato, Luca Iocchi

Auto-TLDR; Low-Complex Neural Networks for Small Data Conditions

Abstract Slides Poster Similar

In this work, we perform a wide variety of experiments with different Deep Learning architectures in small data conditions. We show that model complexity is a critical factor when only a few samples per class are available. Differently from the literature, we improve the state of the art using low complexity models. We show that standard convolutional neural networks with relatively few parameters are effective in this scenario. In many of our experiments, low complexity models outperform state-of-the-art architectures. Moreover, we propose a novel network that uses an unsupervised loss to regularize its training. Such architecture either improves the results either performs comparably well to low capacity networks. Surprisingly, experiments show that the dynamic data augmentation pipeline is not beneficial in this particular domain. Statically augmenting the dataset might be a promising research direction while dropout maintains its role as a good regularizer.

Energy Minimum Regularization in Continual Learning

Xiaobin Li, Weiqiang Wang

Auto-TLDR; Energy Minimization Regularization for Continuous Learning

Abstract Slides Similar

How to give agents the ability of continuous learning like human and animals is still a challenge. In the regularized continual learning method OWM, the constraint of the model on the energy compression of the learned task is ignored, which results in the poor performance of the method on the dataset with a large number of learning tasks. In this paper, we propose an energy minimization regularization(EMR) method to constrain the energy of learned tasks, providing enough learning space for the following tasks that are not learned, and increasing the capacity of the model to the number of learning tasks. A large number of experiments show that our method can effectively increase the capacity of the model and reduce the sensitivity of the model to the number of tasks and the size of the network.

Pseudo Rehearsal Using Non Photo-Realistic Images

Bhasker Sri Harsha Suri, Kalidas Yeturu

Auto-TLDR; Pseudo-Rehearsing for Catastrophic Forgetting

Abstract Slides Poster Similar

Deep Neural networks forget previously learnt tasks when they are faced with learning new tasks. This is called catastrophic forgetting. Rehearsing the neural network with the training data of the previous task can protect the network from catastrophic forgetting.Since rehearsing requires the storage of entire previous data, Pseudo rehearsal was proposed, where samples belonging to the previous data are generated synthetically for rehearsal. In an image classification setting, while current techniques try to generate synthetic data that is photo-realistic, we demonstrated that Neural networks can be rehearsed on data that is not photo-realistic and still achieve good retention of the previous task. We also demonstrated that forgoing the constraint of having photo realism in the generated data can result in a significant reduction in the consumption of computational and memory resources for pseudo rehearsal.

Sequential Domain Adaptation through Elastic Weight Consolidation for Sentiment Analysis

Avinash Madasu, Anvesh Rao Vijjini

Auto-TLDR; Sequential Domain Adaptation using Elastic Weight Consolidation for Sentiment Analysis

Abstract Slides Poster Similar

Elastic Weight Consolidation (EWC) is a technique used in overcoming catastrophic forgetting between successive tasks trained on a neural network. We use this phenomenon of information sharing between tasks for domain adaptation. Training data for tasks such as sentiment analysis (SA) may not be fairly represented across multiple domains. Domain Adaptation (DA) aims to build algorithms that leverage information from source domains to facilitate performance on an unseen target domain. We propose a model-independent framework - Sequential Domain Adaptation (SDA). SDA draws on EWC for training on successive source domains to move towards a general domain solution, thereby solving the problem of domain adaptation. We test SDA on convolutional, recurrent and attention-based architectures. Our experiments show that the proposed framework enables simple architectures such as CNNs to outperform complex state-of-the-art models in domain adaptation of SA. We further observe the effectiveness of a harder first Anti-Curriculum ordering of source domains leads to maximum performance.

Is the Meta-Learning Idea Able to Improve the Generalization of Deep Neural Networks on the Standard Supervised Learning?

Xiang Deng, Zhongfei Zhang

Auto-TLDR; Meta-learning Based Training of Deep Neural Networks for Few-Shot Learning

Abstract Slides Poster Similar

Substantial efforts have been made on improving the generalization abilities of deep neural networks (DNNs) in order to obtain better performances without introducing more parameters. On the other hand, meta-learning approaches exhibit powerful generalization on new tasks in few-shot learning. Intuitively, few-shot learning is more challenging than the standard supervised learning as each target class only has a very few or no training samples. The natural question that arises is whether the meta-learning idea can be used for improving the generalization of DNNs on the standard supervised learning. In this paper, we propose a novel meta-learning based training procedure (MLTP) for DNNs and demonstrate that the meta-learning idea can indeed improve the generalization abilities of DNNs. MLTP simulates the meta-training process by considering a batch of training samples as a task. The key idea is that the gradient descent step for improving the current task performance should also improve a new task performance, which is ignored by the current standard procedure for training neural networks. MLTP also benefits from all the existing training techniques such as dropout, weight decay, and batch normalization. We evaluate MLTP by training a variety of small and large neural networks on three benchmark datasets, i.e., CIFAR-10, CIFAR-100, and Tiny ImageNet. The experimental results show a consistently improved generalization performance on all the DNNs with different sizes, which verifies the promise of MLTP and demonstrates that the meta-learning idea is indeed able to improve the generalization of DNNs on the standard supervised learning.

Can Data Placement Be Effective for Neural Networks Classification Tasks? Introducing the Orthogonal Loss

Brais Cancela, Veronica Bolon-Canedo, Amparo Alonso-Betanzos

Auto-TLDR; Spatial Placement for Neural Network Training Loss Functions

Abstract Slides Poster Similar

Traditionally, a Neural Network classification training loss function follows the same principle: minimizing the distance between samples that belong to the same class, while maximizing the distance to the other classes. There are no restrictions on the spatial placement of deep features (last layer input). This paper addresses this issue when dealing with Neural Networks, providing a set of loss functions that are able to train a classifier by forcing the deep features to be projected over a predefined orthogonal basis. Experimental results shows that these `data placement' functions can overcome the training accuracy provided by the classic cross-entropy loss function.

The Color Out of Space: Learning Self-Supervised Representations for Earth Observation Imagery

Stefano Vincenzi, Angelo Porrello, Pietro Buzzega, Marco Cipriano, Pietro Fronte, Roberto Cuccu, Carla Ippoliti, Annamaria Conte, Simone Calderara

Auto-TLDR; Satellite Image Representation Learning for Remote Sensing

Abstract Slides Poster Similar

The recent growth in the number of satellite images fosters the development of effective deep-learning techniques for Remote Sensing (RS). However, their full potential is untapped due to the lack of large annotated datasets. Such a problem is usually countered by fine-tuning a feature extractor that is previously trained on the ImageNet dataset. Unfortunately, the domain of natural images differs from the RS one, which hinders the final performance. In this work, we propose to learn meaningful representations from satellite imagery, leveraging its high-dimensionality spectral bands to reconstruct the visible colors. We conduct experiments on land cover classification (BigEarthNet) and West Nile Virus detection, showing that colorization is a solid pretext task for training a feature extractor. Furthermore, we qualitatively observe that guesses based on natural images and colorization rely on different parts of the input. This paves the way to an ensemble model that eventually outperforms both the above-mentioned techniques.

Incrementally Zero-Shot Detection by an Extreme Value Analyzer

Sixiao Zheng, Yanwei Fu, Yanxi Hou

Auto-TLDR; IZSD-EVer: Incremental Zero-Shot Detection for Incremental Learning

Abstract Slides Similar

Human beings not only have the ability of recogniz-ing novel unseen classes, but also can incrementally incorporatethe new classes to existing knowledge preserved. However, thezero-shot learning models assume that all seen classes should beknown beforehand, while incremental learning models cannotrecognize unseen classes. This paper introduces a novel andchallenging task of Incrementally Zero-Shot Detection (IZSD),a practical strategy for both zero-shot learning and class-incremental learning in real-world object detection. An innovativeend-to-end model – IZSD-EVer was proposed to tackle this taskthat requires incrementally detecting new classes and detectingthe classes that have never been seen. Specifically, we proposea novel extreme value analyzer to simultaneously detect objectsfrom old seen, new seen, and unseen classes. Additionally andtechnically, we propose two innovative losses, i.e., background-foreground mean squared error loss alleviating the extremeimbalance of the background and foreground of images, andprojection distance loss aligning the visual space and semanticspaces of old seen classes. Experiments demonstrate the efficacyof our model in detecting objects from both the seen and unseenclasses, outperforming the alternative models on Pascal VOC andMSCOCO datasets.

Iterative Label Improvement: Robust Training by Confidence Based Filtering and Dataset Partitioning

Christian Haase-Schütz, Rainer Stal, Heinz Hertlein, Bernhard Sick

Auto-TLDR; Meta Training and Labelling for Unlabelled Data

Abstract Slides Poster Similar

State-of-the-art, high capacity deep neural networks not only require large amounts of labelled training data, they are also highly susceptible to labelling errors in this data, typically resulting in large efforts and costs and therefore limiting the applicability of deep learning. To alleviate this issue, we propose a novel meta training and labelling scheme that is able to use inexpensive unlabelled data by taking advantage of the generalization power of deep neural networks. We show experimentally that by solely relying on one network architecture and our proposed scheme of combining self-training with pseudolabels, both label quality and resulting model accuracy, can be improved significantly. Our method achieves state-of-the-art results, while being architecture agnostic and therefore broadly applicable. Compared to other methods dealing with erroneous labels, our approach does neither require another network to be trained, nor does it necessarily need an additional, highly accurate reference label set. Instead of removing samples from a labelled set, our technique uses additional sensor data without the need for manual labelling. Furthermore, our approach can be used for semi-supervised learning.

The Effect of Multi-Step Methods on Overestimation in Deep Reinforcement Learning

Lingheng Meng, Rob Gorbet, Dana Kulić

Auto-TLDR; Multi-Step DDPG for Deep Reinforcement Learning

Abstract Slides Poster Similar

Multi-step (also called n-step) methods in reinforcement learning (RL) have been shown to be more efficient than the 1-step method due to faster propagation of the reward signal, both theoretically and empirically, in tasks exploiting tabular representation of the value-function. Recently, research in Deep Reinforcement Learning (DRL) also shows that multi-step methods improve learning speed and final performance in applications where the value-function and policy are represented with deep neural networks. However, there is a lack of understanding about what is actually contributing to the boost of performance. In this work, we analyze the effect of multi-step methods on alleviating the overestimation problem in DRL, where multi-step experiences are sampled from a replay buffer. Specifically building on top of Deep Deterministic Policy Gradient (DDPG), we experiment with Multi-step DDPG (MDDPG), where different step sizes are manually set, and with a variant called Mixed Multi-step DDPG (MMDDPG) where an average over different multi-step backups is used as target Q-value. Empirically, we show that both MDDPG and MMDDPG are significantly less affected by the overestimation problem than DDPG with 1-step backup, which consequently results in better final performance and learning speed. We also discuss the advantages and disadvantages of different ways to do multi-step expansion in order to reduce approximation error, and expose the tradeoff between overestimation and underestimation that underlies offline multi-step methods. Finally, we compare the computational resource needs of TD3 and our proposed methods, since they show comparable final performance and learning speed.

Boundary Optimised Samples Training for Detecting Out-Of-Distribution Images

Luca Marson, Vladimir Li, Atsuto Maki

Auto-TLDR; Boundary Optimised Samples for Out-of-Distribution Input Detection in Deep Convolutional Networks

Abstract Slides Poster Similar

This paper presents a new approach to the problem of detecting out-of-distribution (OOD) inputs in image classifications with deep convolutional networks. We leverage so-called boundary samples to enforce low confidence (maximum softmax probabilities) for inputs far away from the training data. In particular, we propose the boundary optimised samples (named BoS) training algorithm for generating them. Unlike existing approaches, it does not require extra generative adversarial network, but achieves the goal by simply back propagating the gradient of an appropriately designed loss function to the input samples. At the end of the BoS training, all the boundary samples are in principle located on a specific level hypersurface with respect to the designed loss. Our contributions are i) the BoS training as an efficient alternative to generate boundary samples, ii) a robust algorithm therewith to enforce low confidence for OOD samples, and iii) experiments demonstrating improved OOD detection over the baseline. We show the performance using standard datasets for training and different test sets including Fashion MNIST, EMNIST, SVHN, and CIFAR-100, preceded by evaluations with a synthetic 2-dimensional dataset that provide an insight for the new procedure.

Fine-Tuning DARTS for Image Classification

Muhammad Suhaib Tanveer, Umar Karim Khan, Chong Min Kyung

Auto-TLDR; Fine-Tune Neural Architecture Search using Fixed Operations

Abstract Slides Poster Similar

Neural Architecture Search (NAS) has gained attraction due to superior classification performance. Differential Architecture Search (DARTS) is a computationally light method. To limit computational resources DARTS makes numerous approximations. These approximations result in inferior performance. We propose to fine-tune DARTS using fixed operations as these are independent of these approximations. Our method offers a good trade-off between the number of parameters and classification accuracy. Our approach improves the top-1 accuracy on Fashion-MNIST, CompCars and MIO-TCD datasets by 0.56%, 0.50%, and 0.39%, respectively compared to the state-of-the-art approaches. Our approach performs better than DARTS, improving the accuracy by 0.28%, 1.64%, 0.34%, 4.5%, and 3.27% compared to DARTS, on CIFAR-10, CIFAR-100, Fashion-MNIST, CompCars, and MIO-TCD datasets, respectively.

MaxDropout: Deep Neural Network Regularization Based on Maximum Output Values

Claudio Filipi Gonçalves Santos, Danilo Colombo, Mateus Roder, Joao Paulo Papa

Auto-TLDR; MaxDropout: A Regularizer for Deep Neural Networks

Abstract Slides Poster Similar

Different techniques have emerged in the deep learning scenario, such as Convolutional Neural Networks, Deep Belief Networks, and Long Short-Term Memory Networks, to cite a few. In lockstep, regularization methods, which aim to prevent overfitting by penalizing the weight connections, or turning off some units, have been widely studied either. In this paper, we present a novel approach called MaxDropout, a regularizer for deep neural network models that works in a supervised fashion by removing (shutting off) the prominent neurons (i.e., most active) in each hidden layer. The model forces fewer activated units to learn more representative information, thus providing sparsity. Regarding the experiments, we show that it is possible to improve existing neural networks and provide better results in neural networks when Dropout is replaced by MaxDropout. The proposed method was evaluated in image classification, achieving comparable results to existing regularizers, such as Cutout and RandomErasing, also improving the accuracy of neural networks that uses Dropout by replacing the existing layer by MaxDropout.

Generalization Comparison of Deep Neural Networks Via Output Sensitivity

Mahsa Forouzesh, Farnood Salehi, Patrick Thiran

Auto-TLDR; Generalization of Deep Neural Networks using Sensitivity

Abstract Slides Similar

Although recent works have brought some insights into the performance improvement of techniques used in state-of-the-art deep-learning models, more work is needed to understand their generalization properties. We shed light on this matter by linking the loss function to the output's sensitivity to its input. We find a rather strong empirical relation between the output sensitivity and the variance in the bias-variance decomposition of the loss function, which hints on using sensitivity as a metric for comparing the generalization performance of networks, without requiring labeled data. We find that sensitivity is decreased by applying popular methods which improve the generalization performance of the model, such as (1) using a deep network rather than a wide one, (2) adding convolutional layers to baseline classifiers instead of adding fully-connected layers, (3) using batch normalization, dropout and max-pooling, and (4) applying parameter initialization techniques.

Compression Strategies and Space-Conscious Representations for Deep Neural Networks

Giosuè Marinò, Gregorio Ghidoli, Marco Frasca, Dario Malchiodi

Auto-TLDR; Compression of Large Convolutional Neural Networks by Weight Pruning and Quantization

Abstract Slides Poster Similar

Recent advances in deep learning have made available large, powerful convolutional neural networks (CNN) with state-of-the-art performance in several real-world applications. Unfortunately, these large-sized models have millions of parameters, thus they are not deployable on resource-limited platforms (e.g. where RAM is limited). Compression of CNNs thereby becomes a critical problem to achieve memory-efficient and possibly computationally faster model representations. In this paper, we investigate the impact of lossy compression of CNNs by weight pruning and quantization, and lossless weight matrix representations based on source coding. We tested several combinations of these techniques on four benchmark datasets for classification and regression problems, achieving compression rates up to 165 times, while preserving or improving the model performance.

Beyond Cross-Entropy: Learning Highly Separable Feature Distributions for Robust and Accurate Classification

Arslan Ali, Andrea Migliorati, Tiziano Bianchi, Enrico Magli

Auto-TLDR; Gaussian class-conditional simplex loss for adversarial robust multiclass classifiers

Abstract Slides Poster Similar

Deep learning has shown outstanding performance in several applications including image classification. However, deep classifiers are known to be highly vulnerable to adversarial attacks, in that a minor perturbation of the input can easily lead to an error. Providing robustness to adversarial attacks is a very challenging task especially in problems involving a large number of classes, as it typically comes at the expense of an accuracy decrease. In this work, we propose the Gaussian class-conditional simplex (GCCS) loss: a novel approach for training deep robust multiclass classifiers that provides adversarial robustness while at the same time achieving or even surpassing the classification accuracy of state-of-the-art methods. Differently from other frameworks, the proposed method learns a mapping of the input classes onto target distributions in a latent space such that the classes are linearly separable. Instead of maximizing the likelihood of target labels for individual samples, our objective function pushes the network to produce feature distributions yielding high inter-class separation. The mean values of the distributions are centered on the vertices of a simplex such that each class is at the same distance from every other class. We show that the regularization of the latent space based on our approach yields excellent classification accuracy and inherently provides robustness to multiple adversarial attacks, both targeted and untargeted, outperforming state-of-the-art approaches over challenging datasets.

Verifying the Causes of Adversarial Examples

Honglin Li, Yifei Fan, Frieder Ganz, Tony Yezzi, Payam Barnaghi

Auto-TLDR; Exploring the Causes of Adversarial Examples in Neural Networks

Abstract Slides Poster Similar

The robustness of neural networks is challenged by adversarial examples that contain almost imperceptible perturbations to inputs which mislead a classifier to incorrect outputs in high confidence. Limited by the extreme difficulty in examining a high-dimensional image space thoroughly, research on explaining and justifying the causes of adversarial examples falls behind studies on attacks and defenses. In this paper, we present a collection of potential causes of adversarial examples and verify (or partially verify) them through carefully-designed controlled experiments. The major causes of adversarial examples include model linearity, one-sum constraint, and geometry of the categories. To control the effect of those causes, multiple techniques are applied such as $L_2$ normalization, replacement of loss functions, construction of reference datasets, and novel models using multi-layer perceptron probabilistic neural networks (MLP-PNN) and density estimation (DE). Our experiment results show that geometric factors tend to be more direct causes and statistical factors magnify the phenomenon, especially for assigning high prediction confidence. We hope this paper will inspire more studies to rigorously investigate the root causes of adversarial examples, which in turn provide useful guidance on designing more robust models.

Meta Learning Via Learned Loss

Sarah Bechtle, Artem Molchanov, Yevgen Chebotar, Edward Thomas Grefenstette, Ludovic Righetti, Gaurav Sukhatme, Franziska Meier

Auto-TLDR; meta-learning for learning parametric loss functions that generalize across different tasks and model architectures

Abstract Slides Similar

Typically, loss functions, regularization mechanisms and other important aspects of training parametric models are chosen heuristically from a limited set of options. In this paper, we take the first step towards automating this process, with the view of producing models which train faster and more robustly. Concretely, we present a meta-learning method for learning parametric loss functions that can generalize across different tasks and model architectures. We develop a pipeline for “meta-training” such loss functions, targeted at maximizing the performance of the model trained under them. The loss landscape produced by our learned losses significantly improves upon the original task-specific losses in both supervised and reinforcement learning tasks. Furthermore, we show that our meta-learning framework is flexible enough to incorporate additional information at meta-train time. This information shapes the learned loss function such that the environment does not need to provide this information during meta-test time.

Norm Loss: An Efficient yet Effective Regularization Method for Deep Neural Networks

Theodoros Georgiou, Sebastian Schmitt, Thomas Baeck, Wei Chen, Michael Lew

Auto-TLDR; Weight Soft-Regularization with Oblique Manifold for Convolutional Neural Network Training

Abstract Slides Poster Similar

Convolutional neural network training can suffer from diverse issues like exploding or vanishing gradients, scaling-based weight space symmetry and covariant-shift. In order to address these issues, researchers develop weight regularization methods and activation normalization methods. In this work we propose a weight soft-regularization method based on the Oblique manifold. The proposed method uses a loss function which pushes each weight vector to have a norm close to one, i.e. the weight matrix is smoothly steered toward the so-called Oblique manifold. We evaluate our method on the very popular CIFAR-10, CIFAR-100 and ImageNet 2012 datasets using two state-of-the-art architectures, namely the ResNet and wide-ResNet. Our method introduces negligible computational overhead and the results show that it is competitive to the state-of-the-art and in some cases superior to it. Additionally, the results are less sensitive to hyperparameter settings such as batch size and regularization factor.

Learning with Multiplicative Perturbations

Xiulong Yang, Shihao Ji

Auto-TLDR; XAT and xVAT: A Multiplicative Adversarial Training Algorithm for Robust DNN Training

Abstract Slides Poster Similar

Adversarial Training (AT) and Virtual Adversarial Training (VAT) are the regularization techniques that train Deep Neural Networks (DNNs) with adversarial examples generated by adding small but worst-case perturbations to input examples. In this paper, we propose xAT and xVAT, new adversarial training algorithms that generate multiplicative perturbations to input examples for robust training of DNNs. Such perturbations are much more perceptible and interpretable than their additive counterparts exploited by AT and VAT. Furthermore, the multiplicative perturbations can be generated transductively or inductively, while the standard AT and VAT only support a transductive implementation. We conduct a series of experiments that analyze the behavior of the multiplicative perturbations and demonstrate that xAT and xVAT match or outperform state-of-the-art classification accuracies across multiple established benchmarks while being about 30% faster than their additive counterparts. Our source code can be found at https://github.com/sndnyang/xvat

Algorithm Recommendation for Data Streams

Jáder Martins Camboim De Sá, Andre Luis Debiaso Rossi, Gustavo Enrique De Almeida Prado Alves Batista, Luís Paulo Faina Garcia

Auto-TLDR; Meta-Learning for Algorithm Selection in Time-Changing Data Streams

Abstract Slides Poster Similar

In the last decades, many companies are taking advantage of massive data generation at high frequencies through knowledge discovery to identify valuable information. Machine learning techniques can be employed for knowledge discovery, since they are able to extract patterns from data and induce models to predict future events. However, dynamic and evolving environments generate streams of data that usually are non-stationary. Models induced in these scenarios may perish over time due to seasonality or concept drift. The periodic retraining could help but the fixed algorithm's hypothesis space could no longer be appropriate. An alternative solution is to use meta-learning for periodic algorithm selection in time-changing environments, choosing the bias that best suits the current data. In this paper, we present an enhanced framework for data streams algorithm selection based on MetaStream. Our approach uses meta-learning and incremental learning to actively select the best algorithm for the current concept in a time-changing. Different from previous works, a set of cutting edge meta-features and an incremental learning approach in the meta-level based on LightGBM are used. The results show that this new strategy can improve the recommendation of the best algorithm more accurately in time-changing data.

Neuron-Based Network Pruning Based on Majority Voting

Ali Alqahtani, Xianghua Xie, Ehab Essa, Mark W. Jones

Auto-TLDR; Large-Scale Neural Network Pruning using Majority Voting

Abstract Slides Poster Similar

The achievement of neural networks in a variety of applications is accompanied by a dramatic increase in computational costs and memory requirements. In this paper, we propose an efficient method to simultaneously identify the critical neurons and prune the model during training without involving any pre-training or fine-tuning procedures. Unlike existing methods, which accomplish this task in a greedy fashion, we propose a majority voting technique to compare the activation values among neurons and assign a voting score to quantitatively evaluate their importance.This mechanism helps to effectively reduce model complexity by eliminating the less influential neurons and aims to determine a subset of the whole model that can represent the reference model with much fewer parameters within the training process. Experimental results show that majority voting efficiently compresses the network with no drop in model accuracy, pruning more than 79\% of the original model parameters on CIFAR10 and more than 91\% of the original parameters on MNIST. Moreover, we show that with our proposed method, sparse models can be further pruned into even smaller models by removing more than 60\% of the parameters, whilst preserving the reference model accuracy.

Knowledge Distillation with a Precise Teacher and Prediction with Abstention

Xu Yi, Jian Pu, Hui Zhao

Auto-TLDR; Knowledge Distillation using Deep gambler loss and selective classification framework

Abstract Slides Poster Similar

Knowledge distillation, which aims to train model under the supervision from another large model (teacher model) to the original model (student model), has achieved remarkable results in supervised learning. However, there are two major problems with existing knowledge distillation methods. One is the teacher's supervision is sometimes misleading, and the other is the student's prediction is not accurate enough. To address the first issue, instead of learning a combination of both teachers and ground truth, we apply knowledge adjustment to correct teachers' supervision using ground truth. For the second problem, we use the selective classification framework to train the student model. In particular, the deep gambler loss is adopted to predict with reservation by explicitly introducing the $(m+1)$-th class. We consider two settings of knowledge distillation: (1) distillation across different network structures ({\it AlexNet, ResNet}), and (2) distillation across networks with different depths ({\it ResNet18, ResNet50}) to evaluate the effectiveness of our method. The experimental results on benchmark datasets (i.e., {\it Fashion-MNIST, SVHN, CIFAR10, CIFAR100}) are reported with higher prediction accuracies and lower coverage errors.

Generative Latent Implicit Conditional Optimization When Learning from Small Sample

Idan Azuri, Daphna Weinshall

Auto-TLDR; GLICO: Generative Latent Implicit Conditional Optimization for Small Sample Learning

Abstract Slides Poster Similar

We revisit the long-standing problem of learning from small sample. The generation of new samples from a small training set of labeled points has attracted increased attention in recent years. In this paper, we propose a novel such method called GLICO (Generative Latent Implicit Conditional Optimization). GLICO learns a mapping from the training examples to a latent space and a generator that generates images from vectors in the latent space. Unlike most recent work, which rely on access to large amounts of unlabeled data, GLICO does not require access to any additional data other than the small set of labeled points. In fact, GLICO learns to synthesize completely new samples for every class using as little as 5 or 10 examples per class, with as few as 10 such classes and no data from unknown classes. GLICO is then used to augment the small training set while training a classifier on the small sample. To this end, our proposed method samples the learned latent space using spherical interpolation (slerp) and generates new examples using the trained generator. Empirical results show that the new sampled set is diverse enough, leading to improvement in image classification in comparison with the state of the art when trained on small samples obtained from CIFAR-10, CIFAR-100, and CUB-200.

RNN Training along Locally Optimal Trajectories via Frank-Wolfe Algorithm

Yun Yue, Ming Li, Venkatesh Saligrama, Ziming Zhang

Auto-TLDR; Frank-Wolfe Algorithm for Efficient Training of RNNs

Abstract Slides Poster Similar

We propose a novel and efficient training method for RNNs by iteratively seeking a local minima on the loss surface within a small region, and leverage this directional vector for the update, in an outer-loop. We propose to utilize the Frank-Wolfe (FW) algorithm in this context. Although, FW implicitly involves normalized gradients, which can lead to a slow convergence rate, we develop a novel RNN training method that, surprisingly, even with the additional cost, the overall training cost is empirically observed to be lower than back-propagation. Our method leads to a new Frank-Wolfe method, that is in essence an SGD algorithm with a restart scheme. We prove that under certain conditions our algorithm has a sublinear convergence rate of $O(1/\epsilon)$ for $\epsilon$ error. We then conduct empirical experiments on several benchmark datasets including those that exhibit long-term dependencies, and show significant performance improvement. We also experiment with deep RNN architectures and show efficient training performance. Finally, we demonstrate that our training method is robust to noisy data.

MetaMix: Improved Meta-Learning with Interpolation-based Consistency Regularization

Yangbin Chen, Yun Ma, Tom Ko, Jianping Wang, Qing Li

Auto-TLDR; MetaMix: A Meta-Agnostic Meta-Learning Algorithm for Few-Shot Classification

Abstract Slides Poster Similar

Model-Agnostic Meta-Learning (MAML) and its variants are popular few-shot classification methods. They train an initializer across a variety of sampled learning tasks (also known as episodes) such that the initialized model can adapt quickly to new tasks. However, within each episode, current MAML-based algorithms have limitations in forming generalizable decision boundaries using only a few training examples. In this paper, we propose an approach called MetaMix. It generates virtual examples within each episode to regularize the backbone models. MetaMix can be applied in any of the MAML-based algorithms and learn the decision boundaries which are more generalizable to new tasks. Experiments on the mini-ImageNet, CUB, and FC100 datasets show that MetaMix improves the performance of MAML-based algorithms and achieves the state-of-the-art result when applied in Meta-Transfer Learning.

Learning Sparse Deep Neural Networks Using Efficient Structured Projections on Convex Constraints for Green AI

Michel Barlaud, Frederic Guyard

Auto-TLDR; Constrained Deep Neural Network with Constrained Splitting Projection

Abstract Slides Poster Similar

In recent years, deep neural networks (DNN) have been applied to different domains and achieved dramatic performance improvements over state-of-the-art classical methods. These performances of DNNs were however often obtained with networks containing millions of parameters and which training required heavy computational power. In order to cope with this computational issue a huge literature deals with proximal regularization methods which are time consuming.\\ In this paper, we propose instead a constrained approach. We provide the general framework for our new splitting projection gradient method. Our splitting algorithm iterates a gradient step and a projection on convex sets. We study algorithms for different constraints: the classical $\ell_1$ unstructured constraint and structured constraints such as the nuclear norm, the $\ell_{2,1} $ constraint (Group LASSO). We propose a new $\ell_{1,1} $ structured constraint for which we provide a new projection algorithm We demonstrate the effectiveness of our method on three popular datasets (MNIST, Fashion MNIST and CIFAR). Experiments on these datasets show that our splitting projection method with our new $\ell_{1,1} $ structured constraint provides the best reduction of memory and computational power. Experiments show that fully connected linear DNN are more efficient for green AI.

Improving Batch Normalization with Skewness Reduction for Deep Neural Networks

Pak Lun Kevin Ding, Martin Sarah, Baoxin Li

Auto-TLDR; Batch Normalization with Skewness Reduction

Abstract Slides Poster Similar

Batch Normalization (BN) is a well-known technique used in training deep neural networks. The main idea behind batch normalization is to normalize the features of the layers ($i.e.$, transforming them to have a mean equal to zero and a variance equal to one). Such a procedure encourages the optimization landscape of the loss function to be smoother, and improve the learning of the networks for both speed and performance. In this paper, we demonstrate that the performance of the network can be improved, if the distributions of the features of the output in the same layer are similar. As normalizing based on mean and variance does not necessarily make the features to have the same distribution, we propose a new normalization scheme: Batch Normalization with Skewness Reduction (BNSR). Comparing with other normalization approaches, BNSR transforms not just only the mean and variance, but also the skewness of the data. By tackling this property of a distribution, we are able to make the output distributions of the layers to be further similar. The nonlinearity of BNSR may further improve the expressiveness of the underlying network. Comparisons with other normalization schemes are tested on the CIFAR-100 and ImageNet datasets. Experimental results show that the proposed approach can outperform other state-of-the-arts that are not equipped with BNSR.

Rethinking Deep Active Learning: Using Unlabeled Data at Model Training

Oriane Siméoni, Mateusz Budnik, Yannis Avrithis, Guillaume Gravier

Auto-TLDR; Unlabeled Data for Active Learning

Abstract Slides Poster Similar

Active learning typically focuses on training a model on few labeled examples alone, while unlabeled ones are only used for acquisition. In this work we depart from this setting by using both labeled and unlabeled data during model training across active learning cycles. We do so by using unsupervised feature learning at the beginning of the active learning pipeline and semi-supervised learning at every active learning cycle, on all available data. The former has not been investigated before in active learning, while the study of latter in the context of deep learning is scarce and recent findings are not conclusive with respect to its benefit. Our idea is orthogonal to acquisition strategies by using more data, much like ensemble methods use more models. By systematically evaluating on a number of popular acquisition strategies and datasets, we find that the use of unlabeled data during model training brings a spectacular accuracy improvement in image classification, compared to the differences between acquisition strategies. We thus explore smaller label budgets, even one label per class.

Graph-Based Interpolation of Feature Vectors for Accurate Few-Shot Classification

Yuqing Hu, Vincent Gripon, Stéphane Pateux

Auto-TLDR; Transductive Learning for Few-Shot Classification using Graph Neural Networks

Abstract Slides Poster Similar

In few-shot classification, the aim is to learn models able to discriminate classes using only a small number of labeled examples. In this context, works have proposed to introduce Graph Neural Networks (GNNs) aiming at exploiting the information contained in other samples treated concurrently, what is commonly referred to as the transductive setting in the literature. These GNNs are trained all together with a backbone feature extractor. In this paper, we propose a new method that relies on graphs only to interpolate feature vectors instead, resulting in a transductive learning setting with no additional parameters to train. Our proposed method thus exploits two levels of information: a) transfer features obtained on generic datasets, b) transductive information obtained from other samples to be classified. Using standard few-shot vision classification datasets, we demonstrate its ability to bring significant gains compared to other works.

Contextual Classification Using Self-Supervised Auxiliary Models for Deep Neural Networks

Sebastian Palacio, Philipp Engler, Jörn Hees, Andreas Dengel

Auto-TLDR; Self-Supervised Autogenous Learning for Deep Neural Networks

Abstract Slides Poster Similar

Classification problems solved with deep neural networks (DNNs) typically rely on a closed world paradigm, and optimize over a single objective (e.g., minimization of the cross- entropy loss). This setup dismisses all kinds of supporting signals that can be used to reinforce the existence or absence of particular patterns. The increasing need for models that are interpretable by design makes the inclusion of said contextual signals a crucial necessity. To this end, we introduce the notion of Self-Supervised Autogenous Learning (SSAL). A SSAL objective is realized through one or more additional targets that are derived from the original supervised classification task, following architectural principles found in multi-task learning. SSAL branches impose low-level priors into the optimization process (e.g., grouping). The ability of using SSAL branches during inference, allow models to converge faster, focusing on a richer set of class-relevant features. We equip state-of-the-art DNNs with SSAL objectives and report consistent improvements for all of them on CIFAR100 and Imagenet. We show that SSAL models outperform similar state-of-the-art methods focused on contextual loss functions, auxiliary branches and hierarchical priors.

Iterative Bounding Box Annotation for Object Detection

Bishwo Adhikari, Heikki Juhani Huttunen

Auto-TLDR; Semi-Automatic Bounding Box Annotation for Object Detection in Digital Images

Abstract Slides Poster Similar

Manual annotation of bounding boxes for object detection in digital images is tedious, and time and resource consuming. In this paper, we propose a semi-automatic method for efficient bounding box annotation. The method trains the object detector iteratively on small batches of labeled images and learns to propose bounding boxes for the next batch, after which the human annotator only needs to correct possible errors. We propose an experimental setup for simulating the human actions and use it for comparing different iteration strategies, such as the order in which the data is presented to the annotator. We experiment on our method with three datasets and show that it can reduce the human annotation effort significantly, saving up to 75% of total manual annotation work.

Towards Robust Learning with Different Label Noise Distributions

Diego Ortego, Eric Arazo, Paul Albert, Noel E O'Connor, Kevin Mcguinness

Auto-TLDR; Distribution Robust Pseudo-Labeling with Semi-supervised Learning

Abstract Slides Similar

Noisy labels are an unavoidable consequence of labeling processes and detecting them is an important step towards preventing performance degradations in Convolutional Neural Networks. Discarding noisy labels avoids a harmful memorization, while the associated image content can still be exploited in a semi-supervised learning (SSL) setup. Clean samples are usually identified using the small loss trick, i.e. they exhibit a low loss. However, we show that different noise distributions make the application of this trick less straightforward and propose to continuously relabel all images to reveal a discriminative loss against multiple distributions. SSL is then applied twice, once to improve the clean-noisy detection and again for training the final model. We design an experimental setup based on ImageNet32/64 for better understanding the consequences of representation learning with differing label noise distributions and find that non-uniform out-of-distribution noise better resembles real-world noise and that in most cases intermediate features are not affected by label noise corruption. Experiments in CIFAR-10/100, ImageNet32/64 and WebVision (real-world noise) demonstrate that the proposed label noise Distribution Robust Pseudo-Labeling (DRPL) approach gives substantial improvements over recent state-of-the-art. Code will be made available.

Local Clustering with Mean Teacher for Semi-Supervised Learning

Zexi Chen, Benjamin Dutton, Bharathkumar Ramachandra, Tianfu Wu, Ranga Raju Vatsavai

Auto-TLDR; Local Clustering for Semi-supervised Learning

Abstract Slides Similar

The Mean Teacher (MT) model of Tarvainen and Valpola has shown favorable performance on several semi-supervised benchmark datasets. MT maintains a teacher model's weights as the exponential moving average of a student model's weights and minimizes the divergence between their probability predictions under diverse perturbations of the inputs. However, MT is known to suffer from confirmation bias, that is, reinforcing incorrect teacher model predictions. In this work, we propose a simple yet effective method called Local Clustering (LC) to mitigate the effect of confirmation bias. In MT, each data point is considered independent of other points during training; however, data points are likely to be close to each other in feature space if they share similar features. Motivated by this, we cluster data points locally by minimizing the pairwise distance between neighboring data points in feature space. Combined with a standard classification cross-entropy objective on labeled data points, the misclassified unlabeled data points are pulled towards high-density regions of their correct class with the help of their neighbors, thus improving model performance. We demonstrate on semi-supervised benchmark datasets SVHN and CIFAR-10 that adding our LC loss to MT yields significant improvements compared to MT and performance comparable to the state of the art in semi-supervised learning.

Rethinking Domain Generalization Baselines

Francesco Cappio Borlino, Antonio D'Innocente, Tatiana Tommasi

Auto-TLDR; Style Transfer Data Augmentation for Domain Generalization

Abstract Slides Poster Similar

Despite being very powerful in standard learning settings, deep learning models can be extremely brittle when deployed in scenarios different from those on which they were trained. Domain generalization methods investigate this problem and data augmentation strategies have shown to be helpful tools to increase data variability, supporting model robustness across domains. In our work we focus on style transfer data augmentation and we present how it can be implemented with a simple and inexpensive strategy to improve generalization. Moreover, we analyze the behavior of current state of the art domain generalization methods when integrated with this augmentation solution: our thorough experimental evaluation shows that their original effect almost always disappears with respect to the augmented baseline. This issue open new scenarios for domain generalization research, highlighting the need of novel methods properly able to take advantage of the introduced data variability.

Efficient Online Subclass Knowledge Distillation for Image Classification

Maria Tzelepi, Nikolaos Passalis, Anastasios Tefas

Auto-TLDR; OSKD: Online Subclass Knowledge Distillation

Abstract Slides Poster Similar

Deploying state-of-the-art deep learning models on embedded systems dictates certain storage and computation limitations. During the recent few years Knowledge Distillation (KD) has been recognized as a prominent approach to address this issue. That is, KD has been effectively proposed for training fast and compact deep learning models by transferring knowledge from more complex and powerful models. However, knowledge distillation, in its conventional form, involves multiple stages of training, rendering it a computationally and memory demanding procedure. In this paper, a novel single-stage self knowledge distillation method is proposed, namely Online Subclass Knowledge Distillation (OSKD), that aims at revealing the similarities inside classes, improving the performance of any deep neural model in an online manner. Hence, as opposed to existing online distillation methods, we are able to acquire further knowledge from the model itself, without building multiple identical models or using multiple models to teach each other, rendering the OSKD approach more efficient. The experimental evaluation on two datasets validates that the proposed method improves the classification performance.

WeightAlign: Normalizing Activations by Weight Alignment

Xiangwei Shi, Yunqiang Li, Xin Liu, Jan Van Gemert

Auto-TLDR; WeightAlign: Normalization of Activations without Sample Statistics

Abstract Slides Poster Similar

Batch normalization (BN) allows training very deep networks by normalizing activations by mini-batch sample statistics which renders BN unstable for small batch sizes. Current small-batch solutions such as Instance Norm, Layer Norm, and Group Norm use channel statistics which can be computed even for a single sample. Such methods are less stable than BN as they critically depend on the statistics of a single input sample. To address this problem, we propose a normalization of activation without sample statistics. We present WeightAlign: a method that normalizes the weights by the mean and scaled standard derivation computed within a filter, which normalizes activations without computing any sample statistics. Our proposed method is independent of batch size and stable over a wide range of batch sizes. Because weight statistics are orthogonal to sample statistics, we can directly combine WeightAlign with any method for activation normalization. We experimentally demonstrate these benefits for classification on CIFAR-10, CIFAR-100, ImageNet, for semantic segmentation on PASCAL VOC 2012 and for domain adaptation on Office-31.

Revisiting the Training of Very Deep Neural Networks without Skip Connections

Oyebade Kayode Oyedotun, Abd El Rahman Shabayek, Djamila Aouada, Bjorn Ottersten

Auto-TLDR; Optimization of Very Deep PlainNets without shortcut connections with 'vanishing and exploding units' activations'

Abstract Slides Poster Similar

Deep neural networks (DNNs) with many layers of feature representations yield state-of-the-art results on several difficult learning tasks. However, optimizing very deep DNNs without shortcut connections known as PlainNets, is a notoriously hard problem. Considering the growing interest in this area, this paper investigates holistically two scenarios that plague the training of very deep PlainNets: (1) the relatively popular challenge of 'vanishing and exploding units' activations', and (2) the less investigated 'singularity' problem, which is studied in details in this paper. In contrast to earlier works that study only the saturation and explosion of units' activations in isolation, this paper harmonizes the inconspicuous coexistence of the aforementioned problems for very deep PlainNets. Particularly, we argue that the aforementioned problems would have to be tackled simultaneously for the successful training of very deep PlainNets. Finally, different techniques that can be employed for tackling the optimization problem are discussed, and a specific combination of simple techniques that allows the successful training of PlainNets having up to 100 layers is demonstrated.

Multi-Modal Deep Clustering: Unsupervised Partitioning of Images

Guy Shiran, Daphna Weinshall

Auto-TLDR; Multi-Modal Deep Clustering for Unlabeled Images

Abstract Slides Poster Similar

The clustering of unlabeled raw images is a daunting task, which has recently been approached with some success by deep learning methods. Here we propose an unsupervised clustering framework, which learns a deep neural network in an end-to-end fashion, providing direct cluster assignments of images without additional processing. Multi-Modal Deep Clustering (MMDC), trains a deep network to align its image embeddings with target points sampled from a Gaussian Mixture Model distribution. The cluster assignments are then determined by mixture component association of image embeddings. Simultaneously, the same deep network is trained to solve an additional self-supervised task. This pushes the network to learn more meaningful image representations and stabilizes the training. Experimental results show that MMDC achieves or exceeds state-of-the-art performance on four challenging benchmarks. On natural image datasets we improve on previous results with significant margins of up to 11% absolute accuracy points, yielding an accuracy of 70% on CIFAR-10 and 61% on STL-10.