A Unified Framework for the Joint Development of Eye Movements and Visual Perception by Bert Shi


Rather than explicitly programming a robot, might it be possible to seed a robot with a minimal structure and allow it to learn how to behave in the environment intelligently, much in the same way a baby develops? As a first step towards such a system, we must have models of the development of perception, the robot's internal representation of the environment within the robot based on its sensory input, and the development of behavior, the generation of intelligent actions based upon the perceived environment. Past work has studied these two problems in isolation. However, this ignores the fact that behavior and sensory perception are mutually dependent. Sensory perception drives behavior, but behavior can also influence the development of sensory perception, by altering the statistics of the sensory input. Thus, it is likely that they develop simultaneously. But how should these two learning processes interact? What constraints do we need to put into place to ensure that the learning succeeds in generating intelligent behavior?

I will describe joint work with Jochen Triesch at the Frankfurt Institute of Advanced Study, which addresses these problems by modeling the joint development of visual perception and the control of eye movements[1]. In particular, I will describe how the concept of efficient coding as originally proposed by Barlow[2], when extended to include the contribution of behavior, can be used to enable a robotic system to develop biologically plausible processing modules for image motion perception and stereo disparity perception, as well as control strategies for smooth pursuit and vergence eye movements.


[1] Y. Zhao, C. Rothkopf, J. Triesch and B. E. Shi, “A Unified Model of the Joint Development of Disparity Selectivity and Vergence Control,” presented at the International Conference on Development and Learning, San Diego, CA, USA, Nov. 2012.

[2] H. Barlow, "Possible principles underlying the transformation of sensory messages" in Sensory Communication, MIT Press, 1961.

Nonvolatile Nanoelectronics – A New Paradigm From STT to MeRAM, Nonvolatile Intelligent Systems by Kang Wang

We will give a brief overview of the energy scaling challenge of CMOS. Then I will describe the advantage of collective spin memory devices in terms of nonvolatility, low switching energy, high speed, high endurance, and scalability. These advantages offer a potential of incorporating nonvolatile spin based memory (e.g., spin transfer torque, STT ) with CMOS. We will then present the recent advances and scaling limits of the STT memory as well as new development using the spin Hall Effect to reduce the write current and energy. To further scale down the switching energy, I will describe a new concept of electric-field control of magnetism, or voltage-controlled magnetoelectric (ME) memory (Me-RAM) . Only recently, it is shown to be possible to use electric field to control metallic magnetism. For the latter, we will describe a couple of fundamental mechanisms of voltage control of magnetic moment and direction. With further advances in this area, we anticipate that the energy reduction of several orders of magnitudes beyond that of STT. The integration of magnetic devices will reduce the standby leakage of CMOS circuits and thus enable further scaling of CMOS with improved performance. These types of device may be integrated directly on top of front-end processed CMOS using back-end process. They will enables standalone and imbedded applications, making possible new generations of nonvolatile instant-on electronics and other systems. I will discuss the dynamics of the switching as well as additional physical processes in improving the switching process. A potential new paradigm of instant-on, nonvolatile electronic circuits and nano-systems may emerge.

[1] P. K. Amiri, et al., "Low Write-Energy Magnetic Tunnel Junctions for High-Speed Spin-Transfer-Torque MRAM," Ieee Electron Device Letters, vol. 32, pp. 57-59, Jan 2011.

[2] J. G. Alzate, P. Khalili Amiri, P. Upadhyaya, S.S. Cherepov, J. Zhu2, M. Lewis, J. A. Katine, J. Langer, K. Galatsis, I. Krivorotov, and K. L. Wang, “Voltage-Induced Switching of Nanoscale Magnetic Tunnel Junctions”, IEDM. 2012.

[3] K. L. Wang and P. Khalilli Amiri, “Nonvolatile Spintronics: Perspectives on Instant-on Nonvolatile Nanoelectronics Systems”, J Spin, Vol. 2, No. 2 (2012) 1250009

Cognitive Computing with Emerging Nanodevices: Material Point of View by Doo Seok Jeong

Abstract This talk includes nanoionics-based electronic transport behaviour and related electronic devices, anion- and cation-migration-induced resistive memories, popularly referred to as resistive random access memories. In his talk he discusses nanoionics thin film materials for artificial neurons and chemical synapses, and implementation of neural functionalities such as spike firing, short- and long-term memories at the single nerve cell level by means of point-defect-migration-dynamics in nanoionic systems.

Challenges and opportunities of ReRAM for neuromorphic engineering by Themis Prodromakis

Abstract This talk is on the challenges and opportunities of ReRAM (Redox Random Access Memory) for neuromorphic engineering. A particular challenge is the co-existence of unipolar/bipolar switching and the volatility effects that solid-state memristors exhibit. He discusses the latest work of his group on how to turn the above mentioned aspects to our benefit in designing cognitive computing based on nanoelectronics.

Dealing with Unknown Unknowns (in Speech Recognition) by Hynek Hermansky