We explore building generative neural network models of popular reinforcement learning environments. Our world model can be trained quickly in an unsupervised manner to learn a compressed spatial and temporal representation of the environment. By using features extracted from the world model as inputs to an agent, we can train a very compact and simple policy that can solve the required task. We can even train our agent entirely inside of its own hallucinated dream generated by its world model, and transfer this policy back into the actual environment. Read More
Tag Archives: GANS
Connecting Touch and Vision via Cross-Modal Prediction
Humans perceive the world using multi-modal sensory inputs such as vision, audition, and touch. In this work, we investigate the cross-modal connection between vision and touch. The main challenge in this cross-domain modeling task lies in the significant scale discrepancy between the two: while our eyes perceive an entire visual scene at once, humans can only feel a small region of an object at any given moment. To connect vision and touch, we introduce new tasks of synthesizing plausible tactile signals from visual inputs as well as imagining how we interact with objects given tactile data as input. To accomplish our goals, we first equip robots with both visual and tactile sensors and collect a large-scale dataset of corresponding vision and tactile image sequences. To close the scale gap, we present a new conditional adversarial model that incorporates the scale and location information of the touch. Human perceptual studies demonstrate that our model can produce realistic visual images from tactile data and vice versa. Finally, we present both qualitative and quantitative experimental results regarding different system designs, as well as visualizing the learned representations of our model. Read More
MIT CSAIL’s AI can visualize objects using touch
Robots that can learn to see by touch are within reach, claim researchers at MIT’s Computer Science and Artificial Intelligence Laboratory. Really. In a newly published paper that’ll be presented next week at the Conference on Computer Vision and Pattern Recognition in Long Beach, California, they describe an AI system capable of generating visual representations of objects from tactile signals, and of predicting tactility from snippets of visual data.
“By looking at the scene, our model can imagine the feeling of touching a flat surface or a sharp edge,” said CSAIL PhD student and lead author on the research Yunzhu Li, who wrote the paper alongside MIT professors Russ Tedrake and Antonio Torralba and MIT postdoc Jun-Yan Zhu. “By blindly touching around, our [AI] model can predict the interaction with the environment purely from tactile feelings. Bringing these two senses together could empower the robot and reduce the data we might need for tasks involving manipulating and grasping objects.” Read More
How the Artificial-Intelligence Program AlphaZero Mastered Its Games
A few weeks ago, a group of researchers from Google’s artificial-intelligence subsidiary, DeepMind, published a paper in the journal Science that described an A.I. for playing games. While their system is general-purpose enough to work for many two-person games, the researchers had adapted it specifically for Go, chess, and shogi (“Japanese chess”); it was given no knowledge beyond the rules of each game. At first it made random moves. Then it started learning through self-play. Over the course of nine hours, the chess version of the program played forty-four million games against itself on a massive cluster of specialized Google hardware. After two hours, it began performing better than human players; after four, it was beating the best chess engine in the world.The best of The New Yorker, in your in-boxReporting, commentary, culture, and humor. Sign up for our newsletters now.
The program, called AlphaZero, descends from AlphaGo, an A.I. that became known for defeating Lee Sedol, the world’s best Go player, in March of 2016. Sedol’s defeat was a stunning upset. In “AlphaGo,” a documentary released earlier this year on Netflix, the filmmakers follow both the team that developed the A.I. and its human opponents, who have devoted their lives to the game. We watch as these humans experience the stages of a new kind of grief. At first, they don’t see how they can lose to a machine: “I believe that human intuition is still too advanced for A.I. to have caught up,” Sedol says, the day before his five-game match with AlphaGo. Then, when the machine starts winning, a kind of panic sets in. In one particularly poignant moment, Sedol, under pressure after having lost his first game, gets up from the table and, leaving his clock running, walks outside for a cigarette. He looks out over the rooftops of Seoul. (On the Internet, more than fifty million people were watching the match.) Meanwhile, the A.I., unaware that its opponent has gone anywhere, plays a move that commentators called creative, surprising, and beautiful. In the end, Sedol lost, 1-4. Before there could be acceptance, there was depression. “I want to apologize for being so powerless,” he said in a press conference. Eventually, Sedol, along with the rest of the Go community, came to appreciate the machine. “I think this will bring a new paradigm to Go,” he said. Fan Hui, the European champion, agreed. “Maybe it can show humans something we’ve never discovered. Maybe it’s beautiful.” Read More
Top GAN Research Papers Every Machine Learning Enthusiast Must Peruse
In the early 1960s, AI pioneer Herbert Simon observed that in a span of two decades, machines will match the cognitive abilities of humankind. Predictions like these motivated theorists, sceptics and thinkers from a cross-section of domains to find ways to use computers to perform routine tasks. From Heron’s Automatons in the first century to Google’s Deep Mind in the 21st century, mankind has yearned to make machines more ‘human’.
The latest developments in AI, especially in the applications of Generative Adversarial Networks (GANs), can help researchers tackle the final frontier for replicating human intelligence. With a new paper being released every week, GANs are proving to be a front-runner for achieving the ultimate — AGI.
Here are a few papers that verify the growing popularity of GANs: Read More
These Three Security Trends Are Key to Decentralize Artificial Intelligence
Decentralized artificial intelligence(AI) is one of the most promising trends in the AI space. The hype around decentralized AI has increased lately with the raise on popularity of blockchain technologies. While the value proposition of decentralized AI systems is very clear from a conceptual standpoint, their implementation is full of challenges. Arguably, the biggest challenges of implementing decentralized AI architectures are in the area of security and privacy.
The foundation of decentralized AI systems is an environment in which different parties such as data providers, data scientists and consumers collaborate to create, train and execute AI models without the need of a centralized authority. That type of infrastructure requires to not only establish unbiased trust between the parties but also solve a few security challenges. Let’s take a very simple scenario of a company that wants to create a series of AI models to detect patterns in their sales data. In a decentralized model, the company will publish a series of datasets to a group of data scientists that will collaborate to create different machine learning models. During that process, the data scientists will interact with other parties that will train and regularize the models. Enforcing the privacy of the data as well as the security of the communications between the different parties is essential to enable the creation of AI models in a decentralized manner. Read More
Self-Supervised GANs
If you aren’t familiar with Generative Adversarial Networks (GANs), they are a massively popular generative modeling technique formed by pitting two Deep Neural Networks, a generator and a discriminator, against each other. This adversarial loss has sparked the interest of many Deep Learning and Artificial Intelligence researchers. However, despite the beauty of the GAN formulation and the eye-opening results of the state-of-the-art architectures, GANs are generally very difficult to train. One of the best ways to get better results with GANs are to provide class labels. This is the basis of the conditional-GAN model. This article will show how Self-Supervised Learning can overcome the need for class labels for training GANs and rival the performance of conditional-GAN models.
Before we get into how Self-Supervised Learning improves GANs, we will introduce the concept of Self-Supervised Learning. Compared to the popular families of Supervised and Unsupervised Learning, Self-Supervised is most similar to Unsupervised Learning. Self-Supervised tasks include things such as image colorization, predicting the relative location of extracted patches from an image, or in this case, predicting the rotation angle of an image. These tasks are dubbed “Self-Supervised” because the data lends itself to these surrogate tasks. In this sense, the Self-Supervised tasks take the form of (X, Y) pairs, however, the X,Y pairs are automatically constructed from the dataset itself and do not require human labeling. The paper discussed in this article summarizes Self-Supervised Learning as, “one can make edits to the given image and ask the network to predict the edited part”. This is the basic idea behind Self-Supervised Learning. Read More
Understanding Generative Adversarial Networks (GANs)
Yann LeCun described it as “the most interesting idea in the last 10 years in Machine Learning”. Of course, such a compliment coming from such a prominent researcher in the deep learning area is always a great advertisement for the subject we are talking about! And, indeed, Generative Adversarial Networks (GANs for short) have had a huge success since they were introduced in 2014 by Ian J. Goodfellow and co-authors in the article Generative Adversarial Nets. Read More
A Beginner’s Guide to Generative Adversarial Networks (GANs)
Generative adversarial networks (GANs) are deep neural net architectures comprised of two nets, pitting one against the other (thus the “adversarial”). GANs were introduced in a paper by Ian Goodfellow and other researchers at the University of Montreal, including Yoshua Bengio, in 2014. Referring to GANs, Facebook’s AI research director Yann LeCun called adversarial training “the most interesting idea in the last 10 years in ML.” Read More
Generative Adversarial Nets
A new framework for estimating generative models via an adversarial process, in which we simultaneously train two models: a generative model G that captures the data distribution, and a discriminative model D that estimates the probability that a sample came from the training data rather than G. The training procedure for G is to maximize the probability of D making a mistake. This framework corresponds to a minimax two-player game. Read More
Rick's Cafe AI 