Andrew C. Felch

Accelerating Brain Circuit Simulations of Object Recognition with CELL Processors

Humans outperform computers on many natural tasks including vision. Given the human ability to recognize objects rapidly and almost effortlessly, it is pragmatically sensible to study and attempt to imitate algorithms used by the brain. Analysis of the anatomical structure and physiological operation of brain circuits has led to derivation of novel algorithms that in initial study have successfully addressed issues of known difficulty in visual processing. These algorithms are slow on uni-processor based systems, thwarting attempts to drive real-time robots for behavioral study, but as might be expected of algorithms designed for highly parallel brain architectures, they are intrinsically parallel and lend themselves to efficient implementation across multiple processors. This paper presents an implementation of such parallel algorithms on a CELL processor and further extends it to a low-cost cluster built using the Sony PlayStation 3 (PS3). The paper describes the modeled brain circuitry, derived algorithms, implementation on the PS3, and initial performance evaluation with respect to both speed and visual object recognition efficacy. The results show that a parallel implementation can achieve a 140x performance improvement on a cluster of 3 PS3s, attaining real-time processing delays. More importantly, we show that the improvements scale linearly, or nearly so in practice. These initial findings, while highly promising in their own right, also provide a new platform to enable extended investigation of large scale brain circuit models. Early prototyping of such large scale models has yielded evidence of their efficacy in recognition of time-varying, partially occluded, scale-invariant objects in arbitrary scenes.

Sensor-rich robots driven by real-time brain circuit algorithms

The analysis of particular telencephalic systems has led to derivation of algorithmic statements of their operation, which have grown to include communicating systems from sensory to motor and back. Like the brain circuits from which they are derived, these algorithms (e.g. Granger, 2006 ) perform and learn from experience. Their perception and action capabilities are often initially tested in simulated environments, which are more controllable and repeatable than robot tests, but it is widely recognized that even the most carefully devised simulated environments typically fail to transfer well to real-world settings. Robot testing raises the specter of engineering requirements and programming minutiae, as well as sheer cost, and lack of standardization of robot platforms. For brain-derived learning systems, the primary desideratum of a robot is not that it have advanced pinpoint motor control, nor extensive scripted or preprogrammed behaviors. Rather, if the goal is to study how the robot can acquire new knowledge via actions, sensing results of actions, and incremental learning over time, as children do, then relatively simple motor capabilities will suffice when combined with high-acuity sensors (sight, sound, touch) and powerful onboard processors. The Brainbot platform is an open-source, sensor-rich robot, designed to enable testing of brain-derived perceptual, motor, and learning algorithms in real-world settings. The system is intended to provide an inexpensive yet highly trainable vehicle to broaden the availability of interactive robots for research. The platform is capable of only relatively simple motor tasks, but contains extensive sensors (visual, auditory , tactile ), intended to correspond to crucial basic enabling characteristics for long-term real-world learning. Humans (and animals) missing sensors and limbs can nonetheless function exceedingly well in the world as long as they have intact brains; analogously, Brainbot has reasonable, limited motor function and all necessary sensors to enable it to function at a highly adaptive level: that is, prioritizing sensorimotor learning over unnecessarily complex dexterity. Brainbot s are being tested with brain-circuit algorithms for hierarchical unsupervised and reinforcement learning , to explore perceptual, action, and language learning 4 Sensor-rich robots driven by real-time brain circuit algorithms