Physical construction -- the ability to compose objects, subject to physical dynamics, in order to serve some function -- is fundamental to human intelligence. Here we introduce a suite of challenging physical construction tasks inspired by how children play with blocks, such as matching a target configuration, stacking and attaching blocks to connect objects together, and creating shelter-like structures over target objects. We then examine how a range of modern deep reinforcement learning agents fare on these challenges, and introduce several new approaches which provide superior performance. Our results show that agents which use structured representations (e.g., objects and scene graphs) and structured policies (e.g., object-centric actions) outperform those which use less structured representations, and generalize better beyond their training when asked to reason about larger scenes. Agents which use model-based planning via Monte-Carlo Tree Search also outperform strictly model-free agents in our most challenging construction problems. We conclude that approaches which combine structured representations and reasoning with powerful learning are a key path toward agents that possess rich intuitive physics, scene understanding, and planning.

The science fiction writer Douglas Adams imagined the greatest computer ever built, Deep Thought, programmed to answer the deepest question ever asked: the Great Question of Life, the Universe, and Everything. After 7.5 million years of processing, Deep Thought revealed its answer: Forty-two (1). As artificial intelligence (AI) systems enter every sector of human endeavor--including science, engineering, and health--humanity is confronted by the same conundrum that Adams encapsulated so succinctly: What good is knowing the answer when it is unclear why it is the answer? What good is a black box? In an informal survey of my colleagues in the physical sciences and engineering, the top reason for not using AI methods such as deep learning, voiced by a substantial majority, was that they did not know how to interpret the results.

Artificial intelligence firm DeepMind has tackled games like Go and Starcraft, but now it is turning its attention to more sober affairs: how to solve school-level maths problems. Researchers at the company tasked an AI with teaching itself to solve arithmetic, algebra and probability problems, among others. It didn't do a very good job: when the neural network was tested on a maths exam taken by 16-year-olds in the UK, it got just 14 out of 40 questions correct, or the equivalent of an E grade.

I eagerly attended my 3rd GPU Technology Conference (GTC): "Deep Learning & AI Conference," in Silicon Valley, March 23-26 as a guest of host NVIDIA. GTC has become my favorite tech event of the year due to its highly focused topic areas that align well with my own; data science, machine learning, AI, and deep learning; plus the show has an academic feel that I appreciate. NVIDIA CEO Jensen Huang delivered another one of his patented marathon keynote address which unveiled the company's vision for the upcoming year. The company had to move the keynote's location from the San Jose Convention Center to a very large hall at San Jose State University (complete with pedicabs provided by Kinetica). At around 2 hours and 40 minutes, Huang's seamless and riviting keynotes are masterful with no notes or teleprompter used.

Do you know the answer to the following problem in arithmetic? What is the sum of 1 1 1 1 1 1 1? If you said "seven," you're right. AI researchers from Google's DeepMind this week published research in which they attempted to train neural networks to solve basic problems in arithmetic, algebra and calculus. The kinds of problems on which a high school student would be typically tested.

The algorithm, which is able to match (and exceed) the accuracy of eye specialists with decades of experience, does more than just signpost the state of AI in 2019. It also highlights the priorities of AI developers going forward, as DeepMind have made it clear that'it will not just offer diagnoses, but also be able to explain how it arrived at the conclusion, and how certain it is of the result.' Explainable AI, then, is at the forefront of the latest developments in the field, and a necessary accompaniment as the technology begins to match - and advance beyond - human capability.

Welcome to the future of insurance, as seen through the eyes of Scott, a customer in the year 2030. Upon hopping into the arriving car, Scott decides he wants to drive today and moves the car into "active" mode. Scott's personal assistant maps out a potential route and shares it with his mobility insurer, which immediately responds with an alternate route that has a much lower likelihood of accidents and auto damage as well as the calculated adjustment to his monthly premium. Scott's assistant notifies him that his mobility insurance premium will increase by 4 to 8 percent based on the route he selects and the volume and distribution of other cars on the road. It also alerts him that his life insurance policy, which is now priced on a "pay-as-you-live" basis, will increase by 2 percent for this quarter. The additional amounts are automatically debited from his bank account. When Scott pulls into his destination's parking lot, his car bumps into one of several parking signs.

We apply text analysis approaches for a specialized search engine for 3D CAD models and associated products. The main goals are to distinguish between actual product descriptions and other text on a website, as well as to decide whether a given text is or contains a product name. For this we use paragraph vectors for text classification, a character-level long short-term memory network (LSTM) for a single word classification and an LSTM tagger based on word embeddings for detecting product names within sentences. Despite the need to collect bigger datasets in our specific problem domain, the first results are promising and partially fit for production use.

Success of deep learning techniques have renewed the interest in development of dialogue systems. However, current systems struggle to have consistent long term conversations with the users and fail to build rapport. Topic spotting, the task of automatically inferring the topic of a conversation, has been shown to be helpful in making a dialog system more engaging and efficient. We propose a hierarchical model with self attention for topic spotting. Experiments on the Switchboard corpus show the superior performance of our model over previously proposed techniques for topic spotting and deep models for text classification. Additionally, in contrast to offline processing of dialog, we also analyze the performance of our model in a more realistic setting i.e. in an online setting where the topic is identified in real time as the dialog progresses. Results show that our model is able to generalize even with limited information in the online setting.

In order to enable high-quality decision making and motion planning of intelligent systems such as robotics and autonomous vehicles, accurate probabilistic predictions for surrounding interactive objects is a crucial prerequisite. Although many research studies have been devoted to making predictions on a single entity, it remains an open challenge to forecast future behaviors for multiple interactive agents simultaneously. In this work, we take advantage of the Generative Adversarial Network (GAN) due to its capability of distribution learning and propose a generic multi-agent probabilistic prediction and tracking framework which takes the interactions among multiple entities into account, in which all the entities are treated as a whole. However, since GAN is very hard to train, we make an empirical research and present the relationship between training performance and hyperparameter values with a numerical case study. The results imply that the proposed model can capture both the mean, variance and multi-modalities of the groundtruth distribution. Moreover, we apply the proposed approach to a real-world task of vehicle behavior prediction to demonstrate its effectiveness and accuracy. The results illustrate that the proposed model trained by adversarial learning can achieve a better prediction performance than other state-of-the-art models trained by traditional supervised learning which maximizes the data likelihood. The well-trained model can also be utilized as an implicit proposal distribution for particle filtered based Bayesian state estimation.