Nonparametric Approximation of Inter-Trace returns (NAIT), for fast learning of Atari games
([22.LBF])
Neural Architecture for ISO-Standard Dialogue Act Tagging
[22.MWB]
Retrieval Augmented Classification for Long Tail Visual Recognition
([22.CVPR])
Internal Symmetry Networks - convolutional neural networks
with a novel weight sharing scheme based on representations of symmetry groups
([08.B],
[09.BL])
demonstrated that a simple recurrent network can be
trained to predict a context-sensitive language
[99.CB,
03.CB]
novel methods for analysing the behavior of
neural networks trained to recognize or predict formal languages
[97.BP].
computer art generated through adversarial training between a HERCL generator
and a GAN-style LeNet critic
[19.B,
18.SB].
The aim is not to mimic the style of human artists;
instead the art emerges from a tradeoff between
low algorithmic complexity and the need to fool the critic
into thinking they are real images
[pickartso.com]
Abstract Deep Networks -
a combination of deep neural networks and learning classifier systems
[12.KB,
14.KB]
improved neuroevolution with complexity-based speciation
[19.HB, 16.HB]
evolving plastic neural networks for online learning
[12.CB,
14.CBO]
Dynamically Defined Functions
[06.HBa]
and novel crossover operators
[06.HBb,
05.HB]
for Grammatical Evolution.
Go: pioneered the use of parallel convolutional neural networks for the game of Go in 2008
[B08],
including tree search with separate networks for move selection and board evaluation
[B09].
GPU's were not very poweful in those days, and my players were only at the level of old players like AmiGo from the 1980's.
The techniques I developed include what would today be called SymNets and Pixel Recurrent Convolutional Networks.
Backgammon: trained a
neural network to play Backgammon using a self-play evolution strategy
[98.PB],
and applied the same algorithm to Tron
[99.BSF]
and Simulated Hockey
[98.BS].
Our technique of only fractionally updating the weights of the champ in the direction of the mutant
has recently gained renewed interest from the RL community,
with massive parallelization and application to OpenAI Gym environments.
developed an efficient optimal trajectory planner for multiple mobile robots, using
parametric cubic splines, which was deployed in the
F180 League of Robocup
[03.TBB]