Presentation is loading. Please wait.

Presentation is loading. Please wait.

Rita Pizzi Department of Computer Science University of Milan

Similar presentations


Presentation on theme: "Rita Pizzi Department of Computer Science University of Milan"— Presentation transcript:

1 Rita Pizzi Department of Computer Science University of Milan
Artificial Neural Network Codifies Sensory and Cognitive Events Identifying Chaotic Attractors in EEG Signals Rita Pizzi Department of Computer Science University of Milan

2 BRAIN DYNAMICS HISTORY
Dynamic system is any type  of system that evolves in time and is represented in function of t; The dynamic system is represented in a multidimensional phase state Complex dynamic systems can have chaotic behavior, i.e. an organized (but non periodic) behavior sensible to the initial conditions. A cahotic attractor can be defined: a trajectory of the dynamical system; contained in a finite volume of the phase space.

3 BRAIN DYNAMICS HISTORY
Edward N. Lorenz 1961 Walter J. Freeman 1991 Giulio Tononi Today Small variations in the initial conditions produce large variations in the long term of the behavior of a system. Relationship between olfactory stimulations and cahotic attractors in neural signals An integrated information theory of consciousness IIT explores the relationship between the temporal information integration and the attractor dynamics of metastable states in the corticothalamic complex. Dynamic configurations that constitute non-fixed-point may form an interesting class of systems with high .

4 OUR RESEARCH A novel model of Artificial Neural Network (ITSOM, Inductive Tracing SOM) can detect cahotic attractors in signals It can show self-organization in single signals It can analyze multiple signals together and show forms of coherence between them It can identify attractors with specific codes It attributes identical codes to similar attractors emerging from similar brain states, perceptions and emotions Are these codes labels of qualia ?

5 THE BIONIC CREATURE (2009) A biological neural network is connected to an Artificial Neural Network that decodes the neural signals The hybrid (biological/electronic) brain operates a minirobot

6 THE BIOLOGICAL NEURONS
We used human neural stem cells cultured for 15 days in order to get mature neurons Neurons grow creating a network We cultured cells directly on Multielectrode arrays (MEAs) previously coated with a matrigel substrate.

7 THE HYBRID SYSTEM The cell activity is recorded simultaneously from 64 channels The system can record the cellular activity for a long time without damaging the cultures It is suitable for our experiments that study the dynamical behavior of a whole neural network

8 THE HYBRID SYSTEM We stimulated the network of neurons by means of directional digital patterns, composed by 8x8 bitmaps Each stimulation is followed by a 1s pause: during that, our ITSOM (Inductive Tracing SOM) Artificial Neural Network elaborates the signals

9 RESULTS After training, the minirobot moves according to our commands
The statistical evaluation after the delivery of 25 random patterns shows an accuracy of 80.11% and a precision of 90.50%.

10 SELF-ORGANIZING MAP Self-organizing Map: unsupervised learning
Two layers: one input layer and one competitive (in general bidimensional) layer Each input neuron is connected to all the nodes of the competitive layer Vector quantization: the n- dimensional input is mapped to a k-dimensional output (k<<n)

11 SELF-ORGANIZING MAP Winner-Take-All rule
Distance calculation between the inputs (signals) xi and the nodes with weights Wvi. The winning node is the neuron with minimum distance D and is rewarded with a positive increase. Wi (t + 1) = Wi (t) + α(t)(Di (t) - Wi (t)) The procedure is cycled and the final map classifies the input.

12 ITSOM The sequence of the SOM winning nodes tends to repeat itself creating a time series of cahotic attractors. These attractors characterize univocally the input element that produces them. The ITSOM network memorizes the time series of the winning nodes, then analyzes them with a z-score method.

13 Z-SCORE The cumulative scores for each input are normalized following the z standardized variable distribution x = number of wins for each input μ = average of scores σ = standard deviation

14 Z-SCORE Once a threshold τ is fixed , 0<τ<1 , we set
z = 1 for z > τ z = 0 for z ≤ τ So each configuration of winning nodes referring to a specific input is represented by a code composed by zeros and ones. It is immediate to compare these codes and identify similar inputs.

15 ITSOM ATTRACTORS Attractors are labeled with a binary code that identifies them univocally, but the flexibility of the ANN allows to attribute the same codes to similar dynamic events.

16 THE NEW PROJECT Aim of the research is to detect with an AI tool the physiological patterns of sensory and cognitive stimuliations, and show how similar stimuli give rise to chaotic attractors identified with identical or similar codes. We can process both individual signals and many signals simultaneously, highlighting the attractors in which  the corresponding dynamic system is evolving.

17 THE NEW PROJECT In the new study we process signals from a 14 electrodes of the EMOTIV EEG system,  connected to immersive glasses that allow a realistic audiovisual experience. The performances of the Emotiv kit have been evaluated in literature as equal  to or better than normal  clinical EEG headsets

18 THE NEW PROJECT

19 THE NEW PROJECT

20 THE NEW PROJECT A Matlab procedure synchronizes the acquired signals with various sensory experiences  presented in a video.

21 THE NEW PROJECT Examining more signals together, we can detect coherence between signals, highlighting the time course of this form of coherence. We can identify individual attractors with a unique code. It is also possible to quantify these complex dynamic events with many parameters.

22 SELECTED ELECTRODES F7 T8 P7 O1 Location. Frontal lobe,Rostral region of superior frontal gyrus Function, Connectivity. BA9 and BA11 make up prefrontal cortex,Executive functions, Cognitive control Location. Temporal lobe Posterior part contains Wernicke's Area Language comprehension Location. Occipital lobe Includes parts of cuneus, lingual gyrus and the lateral occipital gyrus Visual processing Location.Medial part of occipital lobe Initial site of cortical processing of visual information Organized in orientated columns. GAMMA BAND >30 Hz: implied in consciousness and cognition

23 GAMMA BAND - ELECTRODE: O1 occipital lobe
THE FIRST RESULTS GAMMA BAND - ELECTRODE: O1 occipital lobe DARK YELLOW LIGHT YELLOW PINK RED

24 THE FIRST RESULTS The signal attractors show to be different one from another for different color stimulations and show some similarities for similar colors. Attractors are labeled with a binary code that identifies them univocally, and the flexibility of the ANN allows to attribute the same codes to similar dynamic events. GAMMA BAND: DARK YELLOW-> PINK -> LIGHT YELLOW -> RED ->

25 THE EXPERIMENT 7 subjects
Each stimulus ( 10 sec) is followed by a black stimulus (5 sec)

26 RESULTS

27 RESULTS

28 RESULTS

29 RESULTS

30 CONCLUSIONS The ITSOM analysis shows that similar sensory and cognitive stimulations are identified by similar binary codes. Different sensory and cognitive stimulations have different binary codes.

31 CONCLUSIONS This approach is close to that proposed by G. Tononi to identify qualia, but was never fully expressed up to now due to the lack of a robust quantification and representation method. Among the countless number of possible binary codes we can distinguish different dynamic states with unique codes : we will call them qualia codes.

32 FUTURE DEVELOPMENTS We are experimenting sequences of sensory, cognitive and also emotional stimulations, with and without similarities. We aim to complete the experiment with a higher number of subjects. We are able to calculate the Tononi’s PHI value of the attractors: method to quantify neural correlates of qualia that up to now was never used with real brain signals.

33 References 1. R. Pizzi, M. de Curtis, C. Dickson (2002), Evidence of Chaotic Attractors in Cortical Fast Oscillations Tested by an Artificial Neural Network , in: Soft Computing Applications, ed. Bonarisi Masulli Pasi, Physica Verlag Springer. 2. R. Pizzi, D. Rossetti, G. Cino, D. Marino, A.L. Vescovi, W. Baer (2009). A cultured human neural network operates a robotic actuator. BIOSYSTEMS, vol. 95, p 3. W.J. Freeman, Neurodynamics: An Exploration in Mesoscopic Brain Dynamics. Springer 2000 4. D. Balduzzi, and G. Tononi Integrated information in discrete dynamical systems: motivation and theoretical framework. PLoS Comput. Biol. 4: e 5. G. Tononi, Consciousness as Integrated Information: a Provisional Manifesto. Biol. Bull. 215: 216–242. (December 2008) 6. D. Balduzzi, G. Tononi, Integrated Information in Discrete Dynamical Systems: Motivation and Theoretical Framework. PLoS Computational Biology, 1 June 2008 , Volume 4 , Issue 6. 7. Tononi, Integrated information theory of consciousness: an updated account. G. Tononi Archives Italiennes de Biologie, 150: , 2012 (Chapter on Attractor dynamics in the corticothalamic Complex)


Download ppt "Rita Pizzi Department of Computer Science University of Milan"

Similar presentations


Ads by Google