Two years ago, Ars Technica hosted the online premiere of a weird short film called Sunspring, which was mostly remarkable because its entire script was created by an AI. The film's human cast laughed at odd, computer-generated dialogue and stage direction before performing the results in particularly earnest fashion. That film's production duo, Director Oscar Sharp and AI researcher Ross Goodwin, have returned with another AI-driven experiment that, on its face, looks decidedly worse. Blurry faces, computer-generated dialogue, and awkward scene changes fill out this year's Zone Out, a film created as an entry in the Sci-Fi-London 48-Hour Challenge--meaning, just like last time, it had to be produced in 48 hours and adhere to certain specific prompts. That 48-hour limit is worth minding, because Sharp and Goodwin went one bigger this time: they let their AI system, which they call Benjamin, handle the film's entire production pipeline.

Python continues to take leading positions in solving data science tasks and challenges. Last year we made a blog post overviewing the Python's libraries that proved to be the most helpful at that moment. This year, we expanded our list with new libraries and gave a fresh look to the ones we already talked about, focusing on the updates that have been made during the year. Our selection actually contains more than 20 libraries, as some of them are alternatives to each other and solve the same problem. Therefore we have grouped them as it's difficult to distinguish one particular leader at the moment.

Artificial intelligence can now put itself in someone else's shoes. DeepMind has developed a neural network that taught itself to'imagine' a scene from different viewpoints, based on just a single image. Given a 2D picture of a scene – say, a room with a brick wall, and a brightly coloured sphere and cube on the floor – the neural network can generate a 3D view from a different vantage point, rendering the opposite sides of the objects and altering where shadows fall to maintain the same light source. The system, called the Generative Query Network (GQN), can tease out details from the static images to guess at spatial relationships, including the camera's position. "Imagine you're looking at Mt. Everest, and you move a metre – the mountain doesn't change size, which tells you something about its distance from you,"says Ali Eslami who led the project at Deepmind. "But if you look at a mug, it would change position.

We propose a molecular generative model based on the conditional variational autoencoder for de novo molecular design. It is specialized to control multiple molecular properties simultaneously by imposing them on a latent space. As a proof of concept, we demonstrate that it can be used to generate drug-like molecules with five target properties. We were also able to adjust a single property without changing the others and to manipulate it beyond the range of the dataset.

A long standing problem in visual object categorization is the ability of algorithms to generalize across different testing conditions. The problem has been formalized as a covariate shift among the probability distributions generating the training data (source) and the test data (target) and several domain adaptation methods have been proposed to address this issue. While these approaches have considered the single source-single target scenario, it is plausible to have multiple sources and require adaptation to any possible target domain. This last scenario, named Domain Generalization (DG), is the focus of our work. Differently from previous DG methods which learn domain invariant representations from source data, we design a deep network with multiple domain-specific classifiers, each associated to a source domain. At test time we estimate the probabilities that a target sample belongs to each source domain and exploit them to optimally fuse the classifiers predictions. To further improve the generalization ability of our model, we also introduced a domain agnostic component supporting the final classifier. Experiments on two public benchmarks demonstrate the power of our approach.

Deep Neural Networks are highly over-parameterized and the size of the neural networks can be reduced significantly after training without any decrease in performance. One can clearly see this phenomenon in a wide range of architectures trained for various problems. Weight/channel pruning, distillation, quantization, matrix factorization are some of the main methods one can use to remove the redundancy to come up with smaller and faster models. This work starts with a short informative chapter, where we motivate the pruning idea and provide the necessary notation. In the second chapter, we compare various saliency scores in the context of parameter pruning. Using the insights obtained from this comparison and stating the problems it brings we motivate why pruning units instead of the individual parameters might be a better idea. We propose some set of definitions to quantify and analyze units that don't learn and create any useful information. We propose an efficient way for detecting dead units and use it to select which units to prune. We get 5x model size reduction through unit-wise pruning on MNIST.

How to model distribution of sequential data, including but not limited to speech and human motions, is an important ongoing research problem. It has been demonstrated that model capacity can be significantly enhanced by introducing stochastic latent variables in the hidden states of recurrent neural networks. Simultaneously, WaveNet, equipped with dilated convolutions, achieves astonishing empirical performance in natural speech generation task. In this paper, we combine the ideas from both stochastic latent variables and dilated convolutions, and propose a new architecture to model sequential data, termed as Stochastic WaveNet, where stochastic latent variables are injected into the WaveNet structure. We argue that Stochastic WaveNet enjoys powerful distribution modeling capacity and the advantage of parallel training from dilated convolutions. In order to efficiently infer the posterior distribution of the latent variables, a novel inference network structure is designed based on the characteristics of WaveNet architecture. State-of-the-art performances on benchmark datasets are obtained by Stochastic WaveNet on natural speech modeling and high quality human handwriting samples can be generated as well.

We develop a method for user-controllable semantic image inpainting: Given an arbitrary set of observed pixels, the unobserved pixels can be imputed in a user-controllable range of possibilities, each of which is semantically coherent and locally consistent with the observed pixels. We achieve this using a deep generative model bringing together: an encoder which can encode an arbitrary set of observed pixels, latent variables which are trained to represent disentangled factors of variations, and a bidirectional PixelCNN model. We experimentally demonstrate that our method can generate plausible inpainting results matching the user-specified semantics, but is still coherent with observed pixels. We justify our choices of architecture and training regime through more experiments.

An artificial agent for financial risk and returns' prediction is built with a modular cognitive system comprised of interconnected recurrent neural networks, such that the agent learns to predict the financial returns, and learns to predict the squared deviation around these predicted returns. These two expectations are used to build a volatility-sensitive interval prediction for financial returns, which is evaluated on three major financial indices and shown to be able to predict financial returns with higher than 80% success rate in interval prediction in both training and testing, raising into question the Efficient Market Hypothesis. The agent is introduced as an example of a class of artificial intelligent systems that are equipped with a Modular Networked Learning cognitive system, defined as an integrated networked system of machine learning modules, where each module constitutes a functional unit that is trained for a given specific task that solves a subproblem of a complex main problem expressed as a network of linked subproblems. In the case of neural networks, these systems function as a form of an "artificial brain", where each module is like a specialized brain region comprised of a neural network with a specific architecture.

Tensor networks have found a wide use in a variety of applications in physics and computer science, recently leading to both theoretical insights as well as practical algorithms in machine learning. In this work we explore the connection between tensor networks and probabilistic graphical models, and show that it motivates the definition of generalized tensor networks where information from a tensor can be copied and reused in other parts of the network. We discuss the relationship between generalized tensor network architectures used in quantum physics, such as String-Bond States and Entangled Plaquette States, and architectures commonly used in machine learning. We provide an algorithm to train these networks in a supervised learning context and show that they overcome the limitations of regular tensor networks in higher dimensions, while keeping the computation efficient. A method to combine neural networks and tensor networks as part of a common deep learning architecture is also introduced. We benchmark our algorithm for several generalized tensor network architectures on the task of classifying images and sounds, and show that they outperform previously introduced tensor network algorithms. Some of the models we consider can be realized on a quantum computer and may guide the development of near-term quantum machine learning architectures.