| M.Sc. Indar Sugiarto | ||||||||||
|
|
|||||||||
| Research Interests |
|
| Working Fields |
Systemic Neuroscience and Biomedical Engineering PhD topic: "A New Scheme for Inference and Learning through Massively Parallel Hardware Implementation of Digital Stochastic Circuits" The goal of this research is to investigate various types of inference schemes in the domain of machine learning, especially those derived from graphical models, and to develop new minimalistic inference schemes which can be implemented in hardware, especially on Field-Programmable Gate Arrays (FPGAs). We propose to use graphical models due to their very nature characteristics, which resemble the brain’s operation: they can be used to describe massively parallel distributed systems, where overall performance and result come from concurrently communicating local computations of many individual units. Specifically, this research will extensively explore message passing-based inference and sampling-based inference, and also try to find methods to combine both. This scheme will be used to model brain-style information processing, which is amongst the most promising approaches towards machine intelligence, by emulating the behavior of cognitive systems. There is growing evidence showing that our brain processes information by performing probabilistic computations rather than explicit/exact calculation. Here, we propose to implement digital stochastic circuits in a massively parallel configuration to achieve optimum and efficient implementation for a given network architecture. Developing an efficient algorithm for the network requires appropriate tools for simulation and synthesis. Thus, we will explore our methods on reconfigurable FPGAs, which provide abundant resources of logic fabric that are sufficient for implementing such a reconfigurable computation.
Related Project(s): |