When humans observe a physical system, they can easily locate objects, understand their interactions, and anticipate future behavior, even in settings with complicated and previously unseen interactions. For computers, however, learning such models from videos in an unsupervised fashion is an unsolved research problem. In this paper, we present STOVE, a novel state-space model for videos, which explicitly reasons about objects and their positions, velocities, and interactions. It is constructed by combining an image model and a dynamics model in compositional manner and improves on previous work by reusing the dynamics model for inference, accelerating and regularizing training. STOVE predicts videos with convincing physical behavior over hundreds of timesteps, outperforms previous unsupervised models, and even approaches the performance of supervised baselines. We further demonstrate the strength of our model as a simulator for sample efficient model-based control in a task with heavily interacting objects.

Inspired by chaotic firing of neurons in the brain, we propose ChaosNet -- a novel chaos based artificial neural network architecture for classification tasks. ChaosNet is built using layers of neurons, each of which is a 1D chaotic map known as the Generalized Luroth Series (GLS) which has been shown in earlier works to possess very useful properties for compression, cryptography and for computing XOR and other logical operations. In this work, we design a novel learning algorithm on ChaosNet that exploits the topological transitivity property of the chaotic GLS neurons. The proposed learning algorithm gives consistently good performance accuracy in a number of classification tasks on well known publicly available datasets with very limited training samples. Even with as low as 7 (or fewer) training samples/class (which accounts for less than 0.05% of the total available data), ChaosNet yields performance accuracies in the range 73.89 % - 98.33 %. We demonstrate the robustness of ChaosNet to additive parameter noise and also provide an example implementation of a 2-layer ChaosNet for enhancing classification accuracy. We envisage the development of several other novel learning algorithms on ChaosNet in the near future.

Using deep neural networks that are either invariant or equivariant to permutations in order to learn functions on unordered sets has become prevalent. The most popular, basic models are DeepSets [Zaheer et al. 2017] and PointNet [Qi et al. 2017]. While known to be universal for approximating invariant functions, DeepSets and PointNet are not known to be universal when approximating \emph{equivariant} set functions. On the other hand, several recent equivariant set architectures have been proven equivariant universal [Sannai et al. 2019], [Keriven et al. 2019], however these models either use layers that are not permutation equivariant (in the standard sense) and/or use higher order tensor variables which are less practical. There is, therefore, a gap in understanding the universality of popular equivariant set models versus theoretical ones. In this paper we close this gap by proving that: (i) PointNet is not equivariant universal; and (ii) adding a single linear transmission layer makes PointNet universal. We call this architecture PointNetST and argue it is the simplest permutation equivariant universal model known to date. Another consequence is that DeepSets is universal, and also PointNetSeg, a popular point cloud segmentation network (used eg, in [Qi et al. 2017]) is universal. The key theoretical tool used to prove the above results is an explicit characterization of all permutation equivariant polynomial layers. Lastly, we provide numerical experiments validating the theoretical results and comparing different permutation equivariant models.

--Electromagnetic stimulation of the human brain is a key tool for the neurophysiological characterization and diagnosis of several neurological disorders. Transcranial magnetic stimulation (TMS) is one procedure that is commonly used clinically. However, personalized TMS requires a pipeline for accurate head model generation to provide target-specific stimulation. This process includes intensive segmentation of several head tissues based on magnetic resonance imaging (MRI), which has significant potential for segmentation error, especially for low-contrast tissues. Additionally, a uniform electrical conductivity is assigned to each tissue in the model, which is an unrealistic assumption based on conventional volume conductor modeling. This paper proposes a novel approach to the automatic estimation of electric conductivity in the human head for volume conductor models without anatomical segmentation. A convolutional neural network is designed to estimate personalized electrical conductivity values based on anatomical information obtained from T1-and T2-weighted MRI scans. This approach can avoid the time-consuming process of tissue segmentation and maximize the advantages of position-dependent conductivity assignment based on water content values estimated from MRI intensity values. The computational results of the proposed approach provide similar but smoother electric field results for the brain when compared to conventional approaches. In electromagnetic dosimetry applications, the use of computational models that imitate human anatomy is an essential process [1].

In SCALOR, we achieve scalability with respect to the object density by parallelizing both the propagation and discovery processes, reducing the parallel time complexity per scene image to O (1) from O (N) with N the number of objects in an image. We also observe that the serial object processing in SQAIR based on an RNN not only increases the computation time but also deteriorates discovery performance. To this end, we propose a parallel discovery model with much better discovery capacity and performance. Temporally predicting and detecting trajectories of objects, SCALOR can also be regarded as a generative tracking model. In our experiments, we show that SCALOR can model videos with nearly one hundred moving objects along with complex background on synthetic datasets. Furthermore, we evaluate and demonstrate SCALOR on natural videos as well with tens of objects with complex background. The contribution of this work are: (i) We propose the SCALOR model that significantly improves (two orders of magnitude) the scalability with regard to the the object density. It is applicable to nearly a hundred objects with comparable computation time to SQAIR, which scales only to a few objects.

Dataset distillation is a method for reducing dataset sizes: the goal is to learn a small number of synthetic samples containing all the information of a large dataset. This has several benefits: speeding up model training in deep learning, reducing energy consumption, and reducing required storage space. Currently, each synthetic sample is assigned a single `hard' label, which limits the accuracies that models trained on distilled datasets can achieve. Also, currently dataset distillation can only be used with image data. We propose to simultaneously distill both images and their labels, and thus to assign each synthetic sample a `soft' label (a distribution of labels) rather than a single `hard' label. Our improved algorithm increases accuracy by 2-4% over the original dataset distillation algorithm for several image classification tasks. For example, training a LeNet model with just 10 distilled images (one per class) results in over 96% accuracy on the MNIST data. Using `soft' labels also enables distilled datasets to consist of fewer samples than there are classes as each sample can encode information for more than one class. For example, we show that LeNet achieves almost 92% accuracy on MNIST after being trained on just 5 distilled images. We also propose an extension of the dataset distillation algorithm that allows it to distill sequential datasets including texts. We demonstrate that text distillation outperforms other methods across multiple datasets. For example, we are able to train models to almost their original accuracy on the IMDB sentiment analysis task using just 20 distilled sentences.

A BSTRACT The capability of making interpretable and self-explanatory decisions is essential for developing responsible machine learning systems. In this work, we study the learning to explain problem in the scope of inductive logic programming (ILP). We propose Neural Logic Inductive Learning (NLIL), an efficient differentiable ILP framework that learns first-order logic rules that can explain the patterns in the data. In experiments, compared with the state-of-the-art methods, we find NLIL can search for rules that are x10 times longer while remaining x3 times faster. We also show that NLIL can scale to large image datasets, i.e. The recent years have witnessed the growing success of deep learning models in a wide range of applications. However, these models are also criticized for the lack of interpretability in its behavior and decision making process (Lipton, 2016; Mittelstadt et al., 2019), and for being data-hungry. The ability to explain its decision is essential for developing a responsible and robust decision system (Guidotti et al., 2019). On the other hand, logic programming methods, in the form of first-order logic (FOL), are capable of discovering and representing knowledge in explicit symbolic structure that can be understood and examined by human (Evans & Grefenstette, 2018). In this paper, we investigate the learning to explain problem in the scope of inductive logic programming (ILP) which seeks to learn first-order logic rules that explain the data. Traditional ILP methods (Gal arraga et al., 2015) relies on hard matching and discrete logic for rule search which is not tolerant for ambiguous and noisy data (Evans & Grefenstette, 2018). A number of works are proposed for developing differentiable ILP models that combine the strength of neural and logic-based computation (Y ang et al., 2017; Evans & Grefenstette, 2018; Campero et al., 2018; Rockt aschel & Riedel, 2017; Payani & Fekri, 2019).

Existing dialog systems are all monolingual, where features shared among different languages are rarely explored. In this paper, we introduce a novel multilingual dialogue system. Specifically, we augment the sequence to sequence framework with improved shared-private memory. The shared memory learns common features among different languages and facilitates a cross-lingual transfer to boost dialogue systems, while the private memory is owned by each separate language to capture its unique feature. Experiments conducted on Chinese and English conversation corpora of different scales show that our proposed architecture outperforms the individually learned model with the help of the other language, where the improvement is particularly distinct when the training data is limited.

AshPy is a TensorFlow 2.0 library for (distributed) training, evaluation, model selection, and fast prototyping. It is designed to ease the burden of setting up all the nuances of the architectures built to train complex custom deep learning models. AshPy is a library designed to ease the burden of setting up all the nuances of the architectures built to train complex custom deep learning models. Moreover, it provides already prepared trainers for a classifier model and GAN networks. AshPy it is developed around the concepts of Executor, Context, Metric, and Strategies that represents its foundations.

TL;DR Learn about Deep Learning and create Deep Neural Network model to predict customer churn using TensorFlow.js. First day! You've landed this Data Scientist intern job at a large telecom company. You can't stop dreaming about the Lambos and designer clothes you're going to get once you're a Senior Data Scientist. Even your mom is calling to remind you to put your Ph.D. in Statistics diploma on the wall. This is the life, who cares about that you're in your mid-30s and this is your first job ever.