- 【博客】Understanding Agent Cooperation
We employ deep multi-agent reinforcement learning to model the emergence of cooperation. The new notion of sequential social dilemmas allows us to model how rational agents interact, and arrive at more or less cooperative behaviours depending on the nature of the environment and the agents’ cognitive capacity. The research may enable us to better understand and control the behaviour of complex multi-agent systems such as the economy, traffic, and environmental challenges.
2.【博客】From Bayesian Networks to Neural Networks
3.【论文】Learning to protect communications with adversarial neural cryptography
This paper manages to be both tremendous fun and quite thought-provoking at the same time. If I tell you that the central cast contains Alice, Bob, and Eve, you can probably already guess that we’re going to be talking about cryptography (that or reading the paper title ). But this isn’t cryptography as you know it, and nor is it cryptography intended to actually be used to protect any information – to criticise the paper on that front would be to miss the point in my view. Instead what we get is a really interesting twist on adversarial network training and a further demonstration of the kinds of things that such networks are able to learn.
4.【代码】Train a deep learning net with OpenStreetMap features and satellite imagery
Classify roads and features in satellite imagery, by training neural networks with OpenStreetMap (OSM) data.
- Download a chunk of satellite imagery
- Download OSM data that shows roads/features for that area
- Generate training and evaluation data
- Display predictions of mis-registered roads in OSM data, or display raw predictions of ON/OFF
Running the code is as easy as install Docker, make dev, and run a script.
5.【博客】Adversarially Learned Inference
The adversarially learned inference (ALI) model is a deep directed generative model which jointly learns a generation network and an inference network using an adversarial process. This model constitutes a novel approach to integrating efficient inference with the generative adversarial networks (GAN) framework.
What makes ALI unique is that unlike other approaches to learning inference in deep directed generative models (like variational autoencoders (VAEs)), the objective function involves no explicit reconstruction loop. Instead of focusing on achieving a pixel-perfect reconstruction, ALI tends to produce believable reconstructions with interesting variations, albeit at the expense of making some mistakes in capturing exact object placement, color, style and (in extreme cases) object identity. This is a good thing, because 1) capacity is not wasted to model trivial factors of variation in the input, and 2) the learned features are more or less invariant to these trivial factors of variation, which is what is expected of good feature learning.
These strenghts are showcased via the semi-supervised learning tasks on SVHN and CIFAR10, where ALI achieves a performance competitive with state-of-the-art.