Oscillatory Models of Hippocampal Activity and Memory

1. Oscillatory Models of Hippocampal Activity and Memory Roman Borisyuk University of Plymouth, UK In collaboration with

2. Frank HoppensteadtNew York University

3. Outline • Oscillatory model of Hippocampal Activity • Memorization of sequences of events • Theory of epineuronal memory

4. Publications • Borisyuk R.M. and Hoppensteadt, F. (1998) Memorizing and recalling spatial-temporal patterns in an oscillator model of the hippocampus. Biosystems, v.48, 3-10. • Borisyuk R., Denham M., Denham S. and Hoppensteadt F. (1999) Computational models of predictive and memory-related functions of the hippocampus. Reviews in the Neurosciences, v.10, pp.213-232. • Borisyuk R., Hoppensteadt F. (1999) Oscillatory model of the hippocampus: A study of spatio-temporal patterns of neural activity. Biological Cybernetics, v. 81, no.4, pp 359-371. • Borisyuk R., Denham M., Kazanovich Y., Hoppensteadt F. Vinogradova O. (2000). An Oscillatory Neural Network Model of Sparse Distributed Memory and Novelty Detection. BioSystems, 58:265-272 • Borisyuk R., Denham M., Kazanovich Y., Hoppensteadt F., Vinogradova O., (2001). Oscillatory Model of Novelty Detection. Network: Computation in Neural System, 12: 1-20 • Borisyuk R. and Hoppensteadt F.(2004) A theory of epineuronal memory. Neural Networks, 17:1427-1436.

5. Chain Model of Spatio-Temporal Activity We model activity of the hippocampus by a chain of interactive oscillators corresponding to lamellas. Each oscillator has two theta modulated inputs with time shift which controls resulting activity pattern (hippocampal bar code). System demonstrates a wide variety of dynamics: synchronization, non-linear resonance, chaotic activity, etc. Borisyuk & Hoppensteadt, 1999, Biological Cybernetics

6. Model Description We study this model analytically using VCONs (Hoppeansteadt, 1975) and computationally using W-C oscillator (Wilson & Cowan, 1972). En(t) and In(t) are average activities of excitatory and inhibitory populations; Z() is sigmoid; Rn and Vn describe interactions with neighbours; Pn and Qn are periodic inputs D=fC –fS controls patterns of activity

7. Gamma and Theta Rhythms of Single Oscillator Single oscillator under influences of two inputs can demonstrate complex behaviour with slow (theta) and fast (gamma) components Recoding from hippocampal population (Van Quyen & Bragin, 2007) • 0 200 400 t

8. TIME TIME Input: Input: SEPTAL EC Spatio-Temporal Patterns Hippocampal Bar Code D=18 D=5 HIPPOCAMPUS Phase deviation D is a key parameter which coltrols dynamics of hippocampal activity Borisyuk & Hoppensteadt, 1999, Biological Cybernetics

9. Phase/Frequency Coding and Novelty Detection Equations of ONN dynamics: Borisyuk, Denham, Kazanovich, Hoppensteadt, Vinogradova (2000,2001)

10. Model Description Dynamics of oscillator’ frequencies is governed by the learning rule: here we do not modify connection strengths, instead we adjust natural frequency =5 =7 =8

11. Dynamics of Frequencies and Amplitudes Resonant state Non-Resonant state

12. Novelty Detection: Sparse Coding Example of sparse coding: 10 object are coded by 2000 groups The bar’s height is proportional to the number of resonant oscillators in the group. O H E L2 L1 W O D R L The arrow indicates coincidence of resonant oscillator groups for the same symbols “O” 0 2000

13. Oscillatory Memory of Sequences The learning rule is temporally asymmetric, and it takes into account the activity level of pre- and post-”synaptic” neurons in two contiguous time windows. Recall by the network is fast: All memorized patterns of sequences are reproduced in the correct order during the same time window with a short delay. Borisyuk, Denham, Denham, Hoppensteadt (1999)

14. Asymmetric Learning Rule (analog of STDP) w n,j j n Activity Threshold h Time Tm Tm+1 Borisyuk Denham, Denham, Hoppensteadt, 1999, Rev in Neurosc.

18. Oscillatory Memory 0.2 0.4 time 0.6 Example of ONN dynamics. Oscillator consists of 10 excitatory (RED) and 10 inhibitory (BLUE) integrate and fire units with all-to-all connections. The background activity is low. The external input is applied to some group of oscillators during time window. Three time windows are shown.

19. ONN Memory: Sequence of 5 Patterns t t 0.2 0.2 0.4 0.4 0.6 0.6 0.8 0.8 1.0 1.0 1.2 1.2 1.4 1.4 1.6 1.6

20. ONN Memory: Two Sequences t 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2

21. Reverse Replay (Wilson Lab, MIT) Place cell 1 fires Place cell 2 fires Place cell 3 fires Reverse replay Foster & Wilson, Nature, 2006

22. Reverse Replay with Anti STDP

24. Reverse Replay with Anti STDP

25. Forward and Reverse Replay A series of neuronal place-fields, which, when ordered according to the peak in-field firing rates, comprise the place-field sequence “template”. Each neuron’s place-field is shown in a different color. Some sample forward and reverse correlated events from these neurons (same coloring) during immobility. Forward replay Diba & Buzsaki, Nature Neurosc 2007

26. Diba & Buzsaki, Nature Neurosc 2007 Preplay and Replay Forward preplay Reverse replay Spike trains of 13 neurons before, during, and after a single lap (CA1 local field potential shown on top; velocity of the rat shown in the lower panel). The left and right insets magnify 250-ms sections of the spike train, depicting forward preplay and reverse replay, respectively.

27. Epineuronal Memory A theory of epineuronal memory includes a hierarchical structure of variables and parameters that allows us to consider learning and memory processes as being on a variable landscape that is sculptured by reward signals. During fast dynamics, the landscape is attractive quasi-static surface that then slowly guides the system into basin of attraction of the metastable state. A novel mathematical model of Epineuronal Memory is developed that is based on a temporally evolving mnemonic function M, which registers information and guides the dynamics of activity patterns. Borisyuk R & Hoppensteadt F (2004) A theory of epineuronal memory. Neural Networks

28. Formulas of Epineuronal Memory Variables x(t); parameters p(t); stochastic process x(t). Mnemonic landscape function M(t,x,p,x). Reaction-diffusion equation for the landscape function M(t,x,p) Borisyuk R & Hoppensteadt F (2004) A theory of epineuronal memory. Neural Networks

29. Memory of 15 random mnemes REWARD2 REWARD1 REWARD15

30. Recall Starting From Random Initial Data Example of recall Uniform distribution between 15 memorised mnemes. Histogram of 1000 recalls starting from random initial data

31. Recall of Five Sequential Patterns The landscape function peak heights indicate the sequential order of recall

32. Epineuronal Memory: 5 Peaks Mnemonic Surface ZOOM Mnemonic function M(u) 1D vector x x dx/dt Complex dynamics Dynamical uncertainty x

33. Mnemonic Landscape and Trajectories Borisyuk R & Hoppensteadt F (2004) A theory of epineuronal memory. Neural Networks

34. Conclusions • Study of chain model of the hippocampus shows that phase shift between two inputs controls spatio-temporal patterns (hippocampal bar code) • Phase shift, synchronization and resonance have been used to memorise signals and detect their novelty without modification of synaptic strengths • STDP-type learning rule has been used to memorise sequences and replay them in forward and reverse order • General theory of epineuronal memory has been developed which includes both phase-shift and STDP based memories. • The epineuronal paradigm demonstrates mechanisms for stable and persistent memory in the presence of noisy and uncertain environments. It introduces the mnemonic landscape that governs regulation of a brain structures. This approach enables the memorization of events and sequences of events.

35. END PLYMOUTH

36. Happy Birthday to Frank!