AI&BigData Lab 2016. Дмитрий Новицкий: cпайковые и бионические нейронные сети: проблемы и перспекитвы
1 like249 views
1. The document discusses neurons and neural networks from both biological and computational perspectives. It provides historical context on the discovery of neurons in the 17th century and defines key biological concepts like the neuron doctrine, dendrites, axons, and synapses. 2. Functionally, individual neurons are described as input-output devices that integrate excitatory and inhibitory signals at the soma and propagate output signals as spikes along the axon. At the chemical synapse, spike timing influences the strengthening of connections between neurons. 3. Computationally, neural networks are modeled as systems of interconnected neurons that can learn through mechanisms like spike-timing dependent plasticity (STDP) and reinforcement of synaptic transmission based on relative spike timings
AI&BigData Lab 2016. Дмитрий Новицкий: cпайковые и бионические нейронные сети: проблемы и перспекитвы
- 1. Спайковые и бионические нейронные сети: проблемы и перспекитвы Dmitry Nowicki 1
- 3. Нейрон МакКаллока-Питтса (не похож !?) 3
- 4. Historical perspective for “brain ingredients” Key players 1665 – First use of simple microscope to view living cells (Robert Hooke) 1839 – “Cell theory” (Theodor Schwann) – but is it true for the brain? 1870 – Camilo Golgi develops his silver-based method, for randomly staining nerve cells 1887 – S. Ramon Y. Cajal uses Golgi technique – proposes the “neuron doctrine” 1891 – Hienrich Waldeyer – coined the word “Neuron” 1897 - Charles Sherrington coined the word “synapse” Sigmund Freud drawing crayfish neurons, 1882
- 6. Сеть Хопфилда: простейшая ассоциативная память 6 Работает как динамическая система с аттракторами
- 7. Occipital Lobe Parietal Lob Frontal Lobe Temporal Lobe Our “meat machine” what does it consists of?
- 8. Zooming into cortical networks Courtesy of the Blue Brain Team, EPFL
- 9. tical (2-photons) imaging of cortical circuit (anatomy & activity) Courtesy of Adi Mizrahi, Hebrew University
- 10. Adi_Green.jpgAdi_Green.jpg Optical (2-photons) imaging of Hippocampal circuit Courtesy of Adi Mizrahi, Hebrew University
- 12. Layer 5 neocortical pyramidal cells
- 13. The Nobel Prize in Physiology or Medicine 1906 in recognition of their work on the structure of the nervous system Camillo Golgi (1843-1926) Santiago Ramón y Cajal (1852-1934)
- 15. S. Ramon Y Cajal Possible direction of current flow and pattern of axo-dendritic connection The “Neuron Doctrine” and the“Theory of dynamic polarization” axondendrites Dendrites are receptive (input) devices Axon are the sending (output) devices
- 16. Courtesy of Kevan Martin, Univ. Zurich) The neocortical (pyramidal) cell Dendritic tree Axonal tree + varicosities
- 17. The neuron as an input-output electrical device (conceptual, details will follow) Excitatory Inhbitory Soma Axon Dendrites excitatory inhibitory Excitatory post-synaptic potential (EPSPs) spikes Inhibitory post synaptic potential (IPSPs) INPUT OUTPUT
- 18. Spiking activity of a neuron
- 19. Axons “fire” spikes (carrying the brain code) Axonal tree sending output - Spikes
- 20. Typical morphology of a neuron soma dendrites axon nodes of Ranvier Inter-node myelin no myelin Axon terminals (pre-synaptic site) axon initial segment (AIS) “HOT” region generating “spikes” Poliak & Peles Nature Reviews Neuroscience 4, 968-980 (December 2003)
- 21. The myelin of axons Poliak & Peles Nature Reviews Neuroscience 4, 968-980 (December 2003) Myelinating glial cells, oligodendrocytes in the central nervous system (CNS) or Schwann cells in the peripheral nervous system (PNS), form the myelin sheath by enwrapping their membrane several times around the axon. 0.3 - 4mm
- 22. Helen C. Lai and Lily Y. Jan Nature Reviews Neuroscience 7, 548-562 (July 2006) The node of Ranvier in axons Hot (excitable) region
- 23. A typical axon in the central nervous system (CNS summary 1. A single, highly branched, thin (mm) process emerging from the soma. Branched locally but may extend far (many centimeters and even meters) away from the soma 2. At the “hot” axon initial segment (AIS) the spike (“action potential”) is initiated and then propagates along the axon 3. Covered with myelin (isolating) lipid sheath, with intermittent small gaps – the nodes of Ranvier (where “hot” – excitable ion channels reside) 4. Decorated with frequent swellings (axonal boutons) – where the neurotransmitter “hides” (the pre-synaptic site) The axon is the output electrical device of neurons, It generates and carries electrical signals called spikes
- 24. Dendrites Purkinje cell (cerebellum) (Courtesy of M. Hausser) Starburst amacrine cell (retina) (Courtesy of W. Denk) CA1 Pyramidal cell (hippocampus) (Courtesy of D. Johnston)
- 25. An example: The layer 5 cortical pyramidal cell (the “psychic” cell by Cajal) Dendritic spines Dendrites with spines Spiny neurons Typical numbers Total dendritic area – 20,000 mm2 Number of dendritic spines/cell – 8,000 Spine area – ~1 mm2 Number of converging inputs (synapses/cell) – 10,000
- 26. Intracellular injection of Lucifer Yellow in fixed cortical tissue Layer II Layer V Human pyramidal neuron from the neocortex Spines spines 20 mm 1 mm Courtesy of Javier DeFelipe, University Madrid
- 27. Neuron types • Classification by anatomical features (“the face” of dendrites and axons) • Classification – functional (e.g., Excitatory (principal) vs. Inhibitory (inter) neurons) • Classification using electrical/spiking activity pattern • Classification using chemical characteristics • Classification using gene expression
- 28. Microcircuit of the Neocortex Z. J. Huang, G. Di Cristo & F. Ango Nature Reviews Neuroscience 8, 673-686 (September 2007) Principal neurons (excitatory) - axon projects to other brain regions Interneurons (inhibitory) – local axonal projection
- 29. Morphometric-based classification of (inhibitory) interneurons DeFelipe et al., Nature Review neuroscience, 2013
- 30. Electrically-based neuron classification (based of spiking patterns) Courtesy of the Blue Brain data-base
- 31. The Chemical Synapse A (chemical/electrical) device that connects axon of neuron A to dendrites of neuron B Dendrites of neuron B Axon of neuron A (note varicosities)
- 32. A chemical synapse formed between axons and dendrites Axonal terminal (pre-synaptic) Dendritic spine (post-synaptic) Synapse (with gap)
- 33. The chemical synapse Axon cell A (small vesicles) Spine: cell “B” SPIKE at axon (digital - “all or none) Excitatory synaptic potential (analog/graded) excitatory synapse axon Spiny dendrite
- 34. The Chemical Synapse When two cells fire together the synapse between them strengthens Cell A Dendrite of Cell BAxon of Cell A Receptors binding neurotransmitter Vesicles containing neurotransmitter molecules
- 35. Chemical synapse a E. Lisman, Sridhar Raghavachari & Richard W. Tsien Nature Reviews Neuroscience 8, 597-609 (August 2007)
- 36. Vesicle quantal release E. Lisman, Sridhar Raghavachari & Richard W. Tsien Nature Reviews Neuroscience 8, 597-609 (August 2007)
- 37. Vesicle quantal release E. Lisman, Sridhar Raghavachari & Richard W. Tsien Nature Reviews Neuroscience 8, 597-609 (August 2007)
- 38. Vesicle quantal release E. Lisman, Sridhar Raghavachari & Richard W. Tsien Nature Reviews Neuroscience 8, 597-609 (August 2007)
- 39. What neurons “see” when embedded in the (cortical) circuit
- 40. L4 Spiny Stellate Cell covered with (excitatory and inhibotory synapses)
- 41. The neuron as an input-output electrical device (SUMMARY after you’ve been learning) Excitatory Inhbitory Soma Axon Dendrites excitatory inhibitory Excitatory post-synaptic potential (EPSPs) spikes Inhibitory post synaptic potential (IPSPs) INPUT OUTPUT
- 43. 43 Spike-Time Dependent Plasticity (STPD) • Spike-Timing Dependent Plasticity (schematic): The STDP function shows the change of synaptic connections as a function of the relative timing of pre- and postsynaptic spikes after 60 spike pairings.
- 44. 44 STPD (Continued) • Basic STDP Model • The weight change Δwj of a synapse from a presynaptic neuron j depends on the relative timing between presynaptic spike arrivals and postsynaptic spikes. Let us name the presynaptic spike arrival times at synapse j by tfj where f=1,2,3,... counts the presynaptic spikes. Similarly, tn with n=1,2,3,... labels the firing times of the postsynaptic neuron. The total weight change Δwj induced by a stimulation protocol with pairs of pre- and postsynaptic spikes is then (Gerstner and al. 1996, Kempter et al. 1999) • Δwj=∑f=1N∑n=1NW(tni−tfj)(1) • where W(dt)=F(dt) denotes one of the STDP functions (also called learning window) • (Song et al. 2000). The parameters A+ and A− may depend on the current value of the synaptic weight wj . The time constants are on the order of τ+=10ms and τ−=10ms
- 45. The STDP can control this robot
- 46. Схема сети
- 50. Thank you for attention! 50