Profile | Research | Website | Faculty Directory

Research Associate

Ph.D. - University of Toronto, 1990

Phone: 301-405-1381 (lab-301-405-2772)
Email: janh@umd.edu

Research Interests:

I study aggression, partly because aggression has such important consequences that it is inherently valuable to study how adult experiences affect an animal’s level of aggressiveness, but also in order to discover the neural mechanisms underlying the regulation of aggression, because this may help us understand the regulation of many motivations. The goal is to discover how a physical information-processing device, the vertebrate brain, can be wired up to produce instincts and learning. Eventually, such knowledge will guide the design of neuromorphic ‘brains’ for autonomous robots.

Our ‘working hypothesis’ about the neural architecture of motivation is called the ‘machine motivation’ model of instincts and learning. Machine motivation models many known phenomena, and also predicts some new phenomena. We test whether these predictions hold biologically by doing experiments on aggression in Siamese fighting fish, Betta splendens.

Our experiments on aggression in Siamese fighting fish have already documented new phenomena, showing how the aggressiveness of Betta is changed by various captive housing conditions (e.g., social isolation, crowding, subliminal stimuli). In an ongoing experiment a student is using social isolation to cause hyper-aggressiveness, then determining whether the hyper-aggressive Betta adjust quickly back to control levels when the Betta are placed in more natural social environments (so far, hyper-aggressiveness seems surprisingly persistent). Planned projects will help us discover what makes aggression (sometimes) rewarding, and why‘ subliminal’ stimuli induce hyper-aggressiveness.

Our studies of the machine motivation model and other neural architectures will now progress more rapidly, as we are finishing a program called NeuralPainter, with which students can create ‘drag and drop’ neural models, and test them. NeuralPainter’s units are neurons and synapses, rather than neuron compartments, which lets students ignore the ‘chemistry’, and focus on the behavior of networks and architectures. NeuralPainter’s completely visual interface lets students click to alter parameters, and then watch how their neural model behaves. Machine motivation was one of the earliest models to propose spike timing dependent synaptic plasticity (STDP), so NeuralPainter has clickable parameters for exploring the behavior of neural architectures that use various types of relative timing synapses. The discovery that STDP is widespread in vertebrate cortex makes this NeuralPainter feature particularly exciting. A Java applet version is being written, so students can use NeuralPainter over the web.

A planned programming project is to integrate the NeuralPainter GUI with Bettasim, an existing virtual fighting fish program. Bettasim will provide the sensing and motor output for NeuralPainter brains, so students can assess the performance of proposed neural architectures in controlling simulated Betta having simulated fights. This is another step toward prototyping neuromorphic controllers for autonomous robots.

Awards and Honors:
- Leverhulme Fellowship, Edinburgh University, 1992
- Sigma Xi