We consider deep networks, trained via stochastic gradient descent to minimize L2 loss, with the training labels perturbed by independent noise at each iteration. We characterize the behavior of the training dynamics near any parameter vector that achieves zero training error, in terms of an implicit regularization term corresponding to the sum over the data points, of the squared L2 norm of the gradient of the model with respect to the parameter vector, evaluated at each data point. We then leverage this general characterization, which holds for networks of any connectivity, width, depth, and choice of activation function, to show that for 2-layer ReLU networks of arbitrary width and L2 loss, when trained on one-dimensional labeled data $(x_1,y_1),\ldots,(x_n,y_n),$ the only stable solutions with zero training error correspond to functions that: 1) are linear over any set of three or more co-linear training points (i.e. the function has no extra "kinks"); and 2) change convexity the minimum number of times that is necessary to fit the training data. Additionally, for 2-layer networks of arbitrary width, with tanh or logistic activations, we show that when trained on a single $d$-dimensional point $(x,y)$ the only stable solutions correspond to networks where the activations of all hidden units at the datapoint, and all weights from the hidden units to the output, take at most two distinct values, or are zero. In this sense, we show that when trained on "simple" data, models corresponding to stable parameters are also "simple"; in short, despite fitting in an over-parameterized regime where the vast majority of expressible functions are complicated and badly behaved, stable parameters reached by training with noise express nearly the "simplest possible" hypothesis consistent with the data. These results shed light on the mystery of why deep networks generalize so well in practice.

Classification using multimodal data arises in many machine learning applications. It is crucial not only to model cross-modal relationship effectively but also to ensure robustness against loss of part of data or modalities. In this paper, we propose a novel deep learning-based multimodal fusion architecture for classification tasks, which guarantees compatibility with any kind of learning models, deals with cross-modal information carefully, and prevents performance degradation due to partial absence of data. We employ two datasets for multimodal classification tasks, build models based on our architecture and other state-of-the-art models, and analyze their performance on various situations. The results show that our architecture outperforms the other multimodal fusion architectures when some parts of data are not available.

Label noise is inherent in many deep learning tasks when the training set becomes large. A typical approach to tackle noisy labels is using robust loss functions. Categorical cross entropy (CCE) is a successful loss function in many applications. However, CCE is also notorious for fitting samples with corrupted labels easily. In contrast, mean absolute error (MAE) is noise-tolerant theoretically, but it generally works much worse than CCE in practice. In this work, we have three main points. First, to explain why MAE generally performs much worse than CCE, we introduce a new understanding of them fundamentally by exposing their intrinsic sample weighting schemes from the perspective of every sample's gradient magnitude with respect to logit vector. Consequently, we find that MAE's differentiation degree over training examples is too small so that informative ones cannot contribute enough against the non-informative during training. Therefore, MAE generally underfits training data when noise rate is high. Second, based on our finding, we propose an improved MAE (IMAE), which inherits MAE's good noise-robustness. Moreover, the differentiation degree over training data points is controllable so that IMAE addresses the underfitting problem of MAE. Third, the effectiveness of IMAE against CCE and MAE is evaluated empirically with extensive experiments, which focus on image classification under synthetic corrupted labels and video retrieval under real noisy labels.

Question Answering (QA) naturally reduces to an entailment problem, namely, verifying whether some text entails the answer to a question. However, for multi-hop QA tasks, which require reasoning with multiple sentences, it remains unclear how best to utilize entailment models pre-trained on large scale datasets such as SNLI, which are based on sentence pairs. We introduce Multee, a general architecture that can effectively use entailment models for multi-hop QA tasks. Multee uses (i) a local module that helps locate important sentences, thereby avoiding distracting information, and (ii) a global module that aggregates information by effectively incorporating importance weights. Importantly, we show that both modules can use entailment functions pre-trained on a large scale NLI datasets. We evaluate performance on MultiRC and OpenBookQA, two multihop QA datasets. When using an entailment function pre-trained on NLI datasets, Multee outperforms QA models trained only on the target QA datasets and the OpenAI transformer models. The code is available at https://github.com/StonyBrookNLP/multee.

Segmentation of pancreas is important for medical image analysis, yet it faces great challenges of class imbalance, background distractions and non-rigid geometrical features. To address these difficulties, we introduce a Deep Q Network(DQN) driven approach with deformable U-Net to accurately segment the pancreas by explicitly interacting with contextual information and extract anisotropic features from pancreas. The DQN based model learns a context-adaptive localization policy to produce a visually tightened and precise localization bounding box of the pancreas. Furthermore, deformable U-Net captures geometry-aware information of pancreas by learning geometrically deformable filters for feature extraction. Experiments on NIH dataset validate the effectiveness of the proposed framework in pancreas segmentation.

Adversarial training is a training scheme designed to counter adversarial attacks by augmenting the training dataset with adversarial examples. Surprisingly, several studies have observed that loss gradients from adversarially trained DNNs are visually more interpretable than those from standard DNNs. Although this phenomenon is interesting, there are only few works that have offered an explanation. In this paper, we attempted to bridge this gap between adversarial robustness and gradient interpretability. To this end, we identified that loss gradients from adversarially trained DNNs align better with human perception because adversarial training restricts gradients closer to the image manifold. We then demonstrated that adversarial training causes loss gradients to be quantitatively meaningful. Finally, we showed that under the adversarial training framework, there exists an empirical tradeoff between test accuracy and loss gradient interpretability and proposed two potential approaches to resolving this tradeoff. Adversarial attack is an imperceptible perturbation to the input image which causes a deep neural network (DNN) to misclassify the perturbed input image with high confidence (Goodfellow et al., 2015).

A Japanese tech start-up is using deep learning to teach a pair of machines a simple job for a human, but a surprisingly tricky task for a robot - cleaning a bedroom. Though it may seem like a basic, albeit tedious, task for a human, robots find this type of job surprisingly complicated. A Japanese tech start-up is using deep learning to teach AI how to deal with disorder and chaos in a child's room. Deep learning is where algorithms, inspired by the human brain, learn from large amounts of data so they're able to perform complex tasks. Some tasks, like welding car chassis in the exact same way day after day, are easy for robots as it is a repetitive process and the machines do not suffer with boredom in the same way as disgruntled employees.

On Wednesday, March 27, the 2018 Turing Award in computing was given to Yoshua Bengio, Geoffrey Hinton and Yann LeCun for their work on deep learning. Deep learning by complex neural networks lies behind the applications that are finally bringing artificial intelligence out of the realm of science fiction into reality. Voice recognition allows you to talk to your robot devices. Image recognition is the key to self-driving cars. But what, exactly, is deep learning? Dozens of articles tell you that it's a complex, multilayered neural network.

Convolution and the convolutional layer are the major building blocks used in convolutional neural networks. A convolution is the simple application of a filter to an input that results in an activation. Repeated application of the same filter to an input results in a map of activations called a feature map, indicating the locations and strength of a detected feature in an input, such as an image. The innovation of convolutional neural networks is the ability to automatically learn a large number of filters in parallel specific to a training dataset under the constraints of a specific predictive modeling problem, such as image classification. The result is highly specific features that can be detected anywhere on input images. In this tutorial, you will discover how convolutions work in the convolutional neural network.

I have always been fascinated with games. The seemingly infinite options available to perform an action under a tight timeline – it's a thrilling experience. So when I read about the incredible algorithms DeepMind was coming up with (like AlphaGo and AlphaStar), I was hooked. I wanted to learn how to make these systems on my own machine. And that led me into the world of deep reinforcement learning (Deep RL).