Cracking the Neural Code for Vision
2 likes445 views
Dr. Stephen G. Odaibo presented a new model for describing the responses of motion-detecting neurons in an opening keynote address. The sinc wavelet model represents how the temporal frequency character of neurons emerges along the visual processing chain from V1 to MT better than other models. It constructs bandpass filters using sinc wavelets in a way that faithfully represents the properties of neurons in the motion cortex. This model provides new insights into how the brain represents visual information related to motion detection.
Cracking the Neural Code for Vision
- 1. Stephen G. Odaibo M.D., M.S.(Math), M.S.(Comp. Sci.) Quantum Lucid Research Laboratories Dubuque, Iowa, USA Opening Keynote Address Global Ophthalmologists Meeting, Osaka Japan May 16th 2016
- 2. I have no financial conflicts of interest to disclose.
- 4. Talk Outline ◦Background and Motivation ◦Objective ◦Method ◦Results ◦Conclusion 2/35
- 6. We see with our brains ◦ Not with our eyes There is a neural code for vision ◦ Each neuron responds in a particular way to features of a visual scene Background Motivation Objective Method Results Conclusion Acknowledgements 4/35
- 7. The neural codes are described by mathematical functions (models) Our research focuses on motion-detecting neurons Here, we present a better model of the responses of motion-detecting neurons Background Motivation Objective Method Results Conclusion Acknowledgements 5/35
- 13. Hubel and Wiesel, 1959 Background Motivation Objective Method Results Conclusion Acknowledgements 11/35
- 14. Time Space Temporal Frequency Spatial Frequency 12/35 Background Motivation Objective Method Results Conclusion Acknowledgements
- 15. Lowpass filter Bandpass filter 13/35 Background Motivation Objective Method Results Conclusion Acknowledgements
- 16. Marceljia, S. Mathematical description of the responses of simple cortical cells, JOSA 70 (1980) 1297-1300 Gabor wavelet = (Gaussian function)*(Sine function) Background Motivation Objective Method Results Conclusion Acknowledgements 14/35
- 17. Background Motivation Objective Method Results Conclusion Acknowledgements 15/35 Marceljia, S. Mathematical description of the responses of simple cortical cells, JOSA 70 (1980) 1297-1300 Gabor wavelet
- 19. Neurons are connected together in chains (network) ◦ Neuronal hierarchy Each neuron’s response is based on its input neurons, and determines the response of its output neuron Background Motivation Objective Method Results Conclusion Acknowledgements 17/35
- 20. Upstream neurons are more likely to be lowpass temporal frequency filters and downstream neurons are more likely to be bandpass temporal frequency filters http://sites.psu.edu/lukelp137/2013/02/07/passion-post-5-jazz-and-the-saxophone Lowpass filter Bandpass filter http://www.tulane.edu/~bfleury/envirobio/Honors%20Web/StraccoViolins/Index.html music123.com Foster et al (1985); DeAngelis et al (1993b); Hawken et al (1996) Background Motivation Objective Method Results Conclusion Acknowledgements 18/35
- 21. V1 MT Background Motivation Objective Method Results Conclusion Acknowledgements 19/35 Histologically: Rockland (1995, 1998), Anderson et al (1998, 2002); Sincich & Horton (2003) TFFG: Foster et al (1985); DeAngelis et al (1993b); Hawken et al (1996)
- 22. To create a response model that faithfully represents the TFFG emergent property along the motion cortex neuron chain MT V1 Background Motivation Objective Method Results Conclusion Acknowledgements 20/35
- 24. Sinc Wavelet= ◦ (Sinc function) x (Gaussian) Fourier transform of Sinc Wavelet= ◦ (Step function) x (Gaussian) Background Motivation Objective Method Results Conclusion Acknowledgements 22/35
- 25. S.G. Odaibo, Society for Neurosci. Abstracts (2014) Background Motivation Objective Method Results Conclusion Acknowledgements 23/35
- 26. Sinc wavelets with varying centers S.G. Odaibo, Society for Neurosci. Abstracts (2014) Background Motivation Objective Method Results Conclusion Acknowledgements 24/35
- 27. Sinc function = sin(x)/x Background Motivation Objective Method Results Conclusion Acknowledgements 25/35
- 29. Construction of a perfect bandpass filter S.G. Odaibo, Society for Neurosci. Abstracts (2014) Background Motivation Objective Method Results Conclusion Acknowledgements 27/35
- 30. Temporal profile (right column) of Sinc wavelet S.G. Odaibo, Society for Neurosci. Abstracts (2014) Background Motivation Objective Method Results Conclusion Acknowledgements 28/35
- 31. Not all linear combinations yield bandpass filters S.G. Odaibo, Society for Neurosci. Abstracts (2014) Background Motivation Objective Method Results Conclusion Acknowledgements 29/35
- 32. Progressive complexity S.G. Odaibo, Society for Neurosci. Abstracts (2014) Background Motivation Objective Method Results Conclusion Acknowledgements 30/35
- 33. Unlike other models, the sinc wavelet model represents the temporal frequency character along the V1 to MT neuron chain We conclude that the sinc wavelet is a better model for describing the receptive fields of neurons in the motion cortex The sinc wavelet model will provide insights into how the brain represents visual information Background Motivation Objective Method Results Conclusion Acknowledgements 31/35
- 34. Ari Rosenberg PhD (U-W Madison) and for helpful discussion on receptive field databases. My wife, Lisa My Family God Background Motivation Objective Method Results Conclusion Acknowledgements 32/35