The choice of initial learning rate can have a profound effect on the performance of deep networks. We present a class of neural networks with solvable training dynamics, and confirm their predictions empirically in practical deep learning settings. The networks exhibit sharply distinct behaviors at small and large learning rates. The two regimes are separated by a phase transition. In the small learning rate phase, training can be understood using the existing theory of infinitely wide neural networks. At large learning rates the model captures qualitatively distinct phenomena, including the convergence of gradient descent dynamics to flatter minima. One key prediction of our model is a narrow range of large, stable learning rates. We find good agreement between our model's predictions and training dynamics in realistic deep learning settings. Furthermore, we find that the optimal performance in such settings is often found in the large learning rate phase. We believe our results shed light on characteristics of models trained at different learning rates. In particular, they fill a gap between existing wide neural network theory, and the nonlinear, large learning rate, training dynamics relevant to practice.

This paper analyzes the robustness of deep learning models in autonomous driving applications and discusses the practical solutions to address that.

Adaptive binarization methodologies threshold the intensity of the pixels with respect to adjacent pixels exploiting the integral images. In turn, the integral images are generally computed optimally using the summed-area-table algorithm (SAT). This document presents a new adaptive binarization technique based on fuzzy integral images through an efficient design of a modified SAT for fuzzy integrals. We define this new methodology as FLAT (Fuzzy Local Adaptive Thresholding). The experimental results show that the proposed methodology have produced an image quality thresholding often better than traditional algorithms and saliency neural networks. We propose a new generalization of the Sugeno and CF 1,2 integrals to improve existing results with an efficient integral image computation. Therefore, these new generalized fuzzy integrals can be used as a tool for grayscale processing in real-time and deep-learning applications. Index Terms: Image Thresholding, Image Processing, Fuzzy Integrals, Aggregation Functions

American Sign Language recognition is a difficult gesture recognition problem, characterized by fast, highly articulate gestures. These are comprised of arm movements with different hand shapes, facial expression and head movements. Among these components, hand shape is the vital, often the most discriminative part of a gesture. In this work, we present an approach for effective learning of hand shape embeddings, which are discriminative for ASL gestures. For hand shape recognition our method uses a mix of manually labelled hand shapes and high confidence predictions to train deep convolutional neural network (CNN). The sequential gesture component is captured by recursive neural network (RNN) trained on the embeddings learned in the first stage. We will demonstrate that higher quality hand shape models can significantly improve the accuracy of final video gesture classification in challenging conditions with variety of speakers, different illumination and significant motion blurr. We compare our model to alternative approaches exploiting different modalities and representations of the data and show improved video gesture recognition accuracy on GMU-ASL51 benchmark dataset

During the onset of a disaster event, filtering relevant information from the social web data is challenging due to its sparse availability and practical limitations in labeling datasets of an ongoing crisis. In this paper, we show that unsupervised domain adaptation through multi-task learning can be a useful framework to leverage data from past crisis events, as well as exploit additional web resources for training efficient information filtering models during an ongoing crisis. We present a novel method to classify relevant tweets during an ongoing crisis without seeing any new examples, using the publicly available dataset of TREC incident streams that provides labeled tweets with 4 relevant classes across 10 different crisis events. Additionally, our method addresses a crucial but missing component from current research in web science for crisis data filtering models: interpretability. Specifically, we first identify a standard single-task attention-based neural network architecture and then construct a customized multi-task architecture for the crisis domain: Multi-Task Domain Adversarial Attention Network. This model consists of dedicated attention layers for each task and a domain classifier for gradient reversal. Evaluation of domain adaptation for crisis events is performed by choosing a target event as the test set and training on the rest. Our results show that the multi-task model outperformed its single-task counterpart and also, training with additional web-resources showed further performance boost. Furthermore, we show that the attention layer can be used as a guide to explain the model predictions by showcasing the words in a tweet that are deemed important in the classification process. Our research aims to pave the way towards a fully unsupervised and interpretable domain adaptation of low-resource crisis web data to aid emergency responders quickly and effectively.

Federated learning is a distributed machine learning approach to privacy preservation and two major technical challenges prevent a wider application of federated learning. One is that federated learning raises high demands on communication, since a large number of model parameters must be transmitted between the server and the clients. The other challenge is that training large machine learning models such as deep neural networks in federated learning requires a large amount of computational resources, which may be unrealistic for edge devices such as mobile phones. The problem becomes worse when deep neural architecture search is to be carried out in federated learning. To address the above challenges, we propose an evolutionary approach to real-time federated neural architecture search that not only optimize the model performance but also reduces the local payload. During the search, a double-sampling technique is introduced, in which for each individual, a randomly sampled sub-model of a master model is transmitted to a number of randomly sampled clients for training without reinitialization. This way, we effectively reduce computational and communication costs required for evolutionary optimization and avoid big performance fluctuations of the local models, making the proposed framework well suited for real-time federated neural architecture search.

Recent advances demonstrate that irregularly wired neural networks from Neural Architecture Search (NAS) and Random Wiring can not only automate the design of deep neural networks but also emit models that outperform previous manual designs. These designs are especially effective while designing neural architectures under hard resource constraints (memory, MACs, . . . ) which highlights the importance of this class of designing neural networks. However, such a move creates complication in the previously streamlined pattern of execution. In fact one of the main challenges is that the order of such nodes in the neural network significantly effects the memory footprint of the intermediate activations. Current compilers do not schedule with regard to activation memory footprint that it significantly increases its peak compared to the optimum, rendering it not applicable for edge devices. To address this standing issue, we present a memory-aware compiler, dubbed SERENITY, that utilizes dynamic programming to find a sequence that finds a schedule with optimal memory footprint. Our solution also comprises of graph rewriting technique that allows further reduction beyond the optimum. As such, SERENITY achieves optimal peak memory, and the graph rewriting technique further improves this resulting in 1.68x improvement with dynamic programming-based scheduler and 1.86x with graph rewriting, against TensorFlow Lite with less than one minute overhead.

Recent literature implements machine learning techniques to assess corporate credit rating based on financial statement reports. In this work, we analyze the performance of four neural network architectures (MLP, CNN, CNN2D, LSTM) in predicting corporate credit rating as issued by Standard and Poor's. We analyze companies from the energy, financial and healthcare sectors in US. The goal of the analysis is to improve application of machine learning algorithms to credit assessment. To this end, we focus on three questions. First, we investigate if the algorithms perform better when using a selected subset of features, or if it is better to allow the algorithms to select features themselves. Second, is the temporal aspect inherent in financial data important for the results obtained by a machine learning algorithm? Third, is there a particular neural network architecture that consistently outperforms others with respect to input features, sectors and holdout set? We create several case studies to answer these questions and analyze the results using ANOVA and multiple comparison testing procedure.

Deep convolutional neural networks (CNNs) trained with logistic and softmax losses have made significant advancement in visual recognition tasks in computer vision. When training data exhibit class imbalances, the class-wise reweighted version of logistic and softmax losses are often used to boost performance of the unweighted version. In this paper, motivated to explain the reweighting mechanism, we explicate the learning property of those two loss functions by analyzing the necessary condition (e.g., gradient equals to zero) after training CNNs to converge to a local minimum. The analysis immediately provides us explanations for understanding (1) quantitative effects of the class-wise reweighting mechanism: deterministic effectiveness for binary classification using logistic loss yet indeterministic for multi-class classification using softmax loss; (2) disadvantage of logistic loss for single-label multi-class classification via one-vs.-all approach, which is due to the averaging effect on predicted probabilities for the negative class (e.g., non-target classes) in the learning process. With the disadvantage and advantage of logistic loss disentangled, we thereafter propose a novel reweighted logistic loss for multi-class classification. Our simple yet effective formulation improves ordinary logistic loss by focusing on learning hard non-target classes (target vs. non-target class in one-vs.-all) and turned out to be competitive with softmax loss. We evaluate our method on several benchmark datasets to demonstrate its effectiveness.

Even though deep learning have shown unmatched performance on various tasks, neural networks has been shown to be vulnerable to small adversarial perturbation of the input which lead to significant performance degradation. In this work we extend the idea of adding independent Gaussian noise to weights and activation during adversarial training (PNI) to injection of colored noise for defense against common white-box and black-box attacks. We show that our approach outperforms PNI and various previous approaches in terms of adversarial accuracy on CIFAR-10 dataset. In addition, we provide an extensive ablation study of the proposed method justifying the chosen configurations.