L. Zhu*, H. Shah*, H. Wang*, M. White

We explore the effectiveness of q-exponential policies in policy optimization methods, finding that heavy-tailed policies (q > 1) are generally more effective and can consistently outperform the Gaussian policy.

K. Javed, H. Shah, R. Sutton, M. White

We show that by either decomposing the network into independent modules or learning a recurrent network incrementally, we can make RTRL scale linearly with the number of parameters. Unlike prior scalable gradient estimation algorithms, our algorithms do not add noise or bias to the gradient estimate.

K. Janjua, H. Shah, M. White, E. Miahi, M. C. Machado, A. White

We propose a framework for making accurate predictions on a real world water treatment plant based on the General Value Functions. This work is one of the first to motivate the importance of adapting predictions in real-time, for non-stationary high-volume systems in the real world.

H. Shah, J. Villmow, A. Ulges, U. Schwanecke, F. Shafait

We propose an extension that enables any existing Knowledge Graph Completion model to predict facts about the open-world entities. This approach is more robust, more portable and has better performance than the published state of the art on most datasets. We also released a new dataset that overcomes the shortcomings of previous ones.

H. Wang, H. Shah, A. White, M. White

We introduce a new fine-tuning algorithm that gradually allows more exploration based on off-policy estimates of performance.

L. Zhu, H. Shah, C. Zheng, N. Yukie, M. White

We study Symmetric Divergences for Behavior Regularized Policy Optimization and propose Symmetric 𝑓-Actor Critic, which avoids per-environment failures encountered by other offline Reinforcement Learning methods.

H. Shah, J. Villmow, A. Ulges

We introduced relation specific transformations to substantially improve the performance of Open World Knowledge Graph Completion models. We also proposed an approach for clustering of relations to reduce the training time and memory footprint.

H. Shah, K. Javed, F. Shafait.

Standard neural networks suffer from catastrophic forgetting when they are trained on incrementally arriving stream of i.i.d. data. To combat this forgetting, one approach is to train GANs on previously arrived data and feed it to the network again. In this paper, we highlighted that this method is biased and proposed an approach to mitigate this bias and reduce the effect of catastrophic forgetting.

A significant proportion of the representations learned by the current generate & test methods consist of highly redundant features. This talk demonstrates how the feature ranking criteria utilized by these methods are highly ineffective in addressing this problem. In this talk, I present a new approach for decorrelating features in an online setting. I demonstrate that this decorrelator can effectively eliminate redundant features and produce a statistically significant performance improvement in the low-capacity function approximation setting.