toward tractable agi: challenges for system identification in neural circuitry
TRANSCRIPT
Toward Tractable AGI: Challenges for Toward Tractable AGI: Challenges for System Identification in Neural CircuitrySystem Identification in Neural Circuitry
Randal A. KoeneCarboncopies.org & NeuraLink Co.
AGI-12, Winter Intelligence Conference 2012, Oxford
Interfacesprostheses
Reconstructionproject
Specialtools
Specific System Identification Problems
Tractable AGI through System Identification in Neural Circuitry
Representations and ModelsBehavior of interest; Signals of interest; Discovering the transfer function
Mental Processes and Neural Circuitry: Brain Emulation
System identification in neural circuitry
Simplification of an Intractable System into a Collection of System Identification Problems
Tools for structural decomposition; Data from structure; Parameter tuning among connected systems; Tools for characteristic reference recordings
ChallengesSignals and predicting spikes; Validation, reconstruction and plasticity; Interference during measurement; Data quantities; Proof of concept
Tractable AGIChallenges in ourenvironment
Legg, Hutter
Optimal boundedlengthspacetime embeddedagent Orseau
Theoretically soundAGI
Practical feasibility
Short-cuts
Brain-like AGI
Abstraction level special case:Neuronal circuitry/physiology(100 years of grounding)
Our reverse interests: Taking a niche systemand making it more adaptable, more general
Tractable AGI through System Identification in Neural Circuitry
Representations and ModelsBehavior of interest; Signals of interest; Discovering the transfer function
Mental Processes and Neural Circuitry: Brain Emulation
System identification in neural circuitry
Simplification of an Intractable System into a Collection of System Identification Problems
Tools for structural decomposition; Data from structure; Parameter tuning among connected systems; Tools for characteristic reference recordings
ChallengesSignals and predicting spikes; Validation, reconstruction and plasticity; Interference during measurement; Data quantities; Proof of concept
Representations and Models
Modern science:
Observed effects
Model (testing)
Understanding
Pieces of natural environment
Not independent!
Signals, information, P(x)
described as
improve
May seem obvious to comp.neurophys. modelers... but consider whole problem not only typical solutions
Behavior of Interest
Lots of piecesSystematic modeling
Keep it simple
Interesting effectFocus, constrain scope/model
Neuroscience:Effect = Behavior(e.g. object recognition)
Signals of Interest
How do pieces communicate?
Signals
Physics: 4 interactions (gravity, electromagnetism, weak & strong nuclear force)
Constrain
Neurons: current, temperature, pressure, EM, etc…
Priority of interest: empirical (noise, predictive value)
Discovering the Transfer FunctionSI in Control Theory: Black/gray box
State, input, output
Find: Transfer Function
Formal methodsE.g. Volterra series expansion
kernels & history of input
Tractable AGI through System Identification in Neural Circuitry
Representations and ModelsBehavior of interest; Signals of interest; Discovering the transfer function
Mental Processes and Neural Circuitry: Brain Emulation
System identification in neural circuitry
Simplification of an Intractable System into a Collection of System Identification Problems
Tools for structural decomposition; Data from structure; Parameter tuning among connected systems; Tools for characteristic reference recordings
ChallengesSignals and predicting spikes; Validation, reconstruction and plasticity; Interference during measurement; Data quantities; Proof of concept
Mental Processes and Neural Circuitry: Brain Emulation
Effects = Experiences
Perception, learning & memory, goal directed decision making, emotional responses, consciousness, language, motor
Observable / internal
Involves Involves ensemblesensembles of of neurons in a circuit neurons in a circuit layoutlayout
Reconstruction vs. Abstraction: Interfaces & Prostheses100 years of component level neuroscience
Individual differences
Matter to interfaces
Matter to prostheses
System Identification in Neural Circuitry
Signals of interestChip – “bits”
Initial assumptions, reliable neural communication
Sensory, muscle, learning – “spikes”
Example methods:Berger chip(Volterra exp.)
Aurel A. Lazar: Channel Identification Machines
(CNS2012 workshop on SI)
Tractable AGI through System Identification in Neural Circuitry
Representations and ModelsBehavior of interest; Signals of interest; Discovering the transfer function
Mental Processes and Neural Circuitry: Brain Emulation
System identification in neural circuitry
Simplification of an Intractable System into a Collection of System Identification Problems
Tools for structural decomposition; Data from structure; Parameter tuning among connected systems; Tools for characteristic reference recordings
ChallengesSignals and predicting spikes; Validation, reconstruction and plasticity; Interference during measurement; Data quantities; Proof of concept
Simplification of an Intractable System into a Collection of System Identification Problems
SI of observable + internal = intractable if black box is brain
Many communicating black boxes with accessible I/O
Communication = note locations, trace connectivity (“Connectomics”)
E.g. compartmental modelingBriggman et al & Bock et al, Nature 2011
Characteristic responsesADP, AHP, AMPA, s/fNMDA
Whole Brain Emulation: A Roadmap to data acquisition & representation
Break into parts.How can the parts communicate?
Characterize the parts
Platform suitingrepresentation
IteratingImproving theOther pillars
(See earlier presentations, carboncopies.org.(See earlier presentations, carboncopies.org.More about roadmap & projects – leading upMore about roadmap & projects – leading upTo Global Future 2045 Congress NYC, June 15-16.)To Global Future 2045 Congress NYC, June 15-16.)
Tools for Structural Decomposition
Voxel geometric decomposition (e.g. MRI)
Cell body locations & functional connectivity
Zador RNA/DNA tags
Stacks of EM images (Denk, Hayworth, Lichtman)
Data from Structure
SI for compartmentsElectric circuit analogy
3D shapeConductance, class, etc.
“Invisible” parameters? Measurement reliability?
Parameter Tuning among Connected Systems: Reference Points
Parameters – sensible collective behavior
Reference points: constrain & validate
Resolution of reference points – combinatorial size of SI problem
# and duration of measurements(purposely abstract:
- resolutions reference/SI decomposition- not one path (e.g. Briggman et al.= problem specific criteria, not method specific – compare & collaborate projects)
Tools for Characteristic Reference RecordingsLarge arrays of recording electrodes + optogenetic selectivity
Microscopic wireless probes
Molecular “ticker-tape” by DNA amplification
Tractable AGI through System Identification in Neural Circuitry
Representations and ModelsBehavior of interest; Signals of interest; Discovering the transfer function
Mental Processes and Neural Circuitry: Brain Emulation
System identification in neural circuitry
Simplification of an Intractable System into a Collection of System Identification Problems
Tools for structural decomposition; Data from structure; Parameter tuning among connected systems; Tools for characteristic reference recordings
ChallengesSignals and predicting spikes; Validation, reconstruction and plasticity; Interference during measurement; Data quantities; Proof of concept
ChallengesGeneral SI problems
Particular to neurons & neural models
Unique to pieces of neural tissue & large neuronal circuits
Exclusive to whole brain circuit reconstruction
Integration of data from structure & function acquisition tools
Care about signalsContributions outside spiking domain?
Other cells?
Neuron-neuron effects without spiking?
Evidence of sensations retained?
Assume: spikes = currency of sensations
Not epiphenomenal! (test?)
Predicting spikes
Observe / deduce spike times original system
Additional information aids prediction
What information do the tools give us?
Izhikevich
MRI
Large volumes
Not parameter tuning
But system validation!Distribution
Propagation
Requires spatial registration
3D reconstructions at 5nmGeneral classification (e.g. pyramidal vs interneuron)
Detailed morphology, segmented into compartments
E.g. radius – resistance, capacitance
Depends on neuron type
Measurement reliability, cumulative
Plasticity & Morphology
Learning changes synapses & connectome
Deformation changes morphology
3D snapshot cannot capture temporal dynamics of memory
Ticker-Tape DataMany neurons, several tapes per neuron
Time stamps + spike / membrane potential samples
Recovery of DNA snippetsNot combinable with EM
Interference with cell mechanisms
Spatial registration:Which part of ultrastructure did it come from?
Optical Functional
Calcium / proteins, fluorescent
Large scale / whole brain access?
Methods disturb tissue
Huge electrode arrays also disturb tissue
Microscopic wirelessPower & data volumes compete with continuous sampling
When enough sporadic data?
Long term dynamicsDemands frequent spatial registration
EM registration in tissue
Ongoing collaboration (MIT, Harvard)
Sufficient Data
Spike times, EFPs, membrane potentials – rate, duration?
Response shape sufficient?
Stimulate combinations?
Learning from virtual systems
NETMORPHAcquire structure data
Acquire functional data
Test algorithms & iteratively improving constraints
Calculate abstract boundary conditions?
Netmorph.org
Proof of Concept: Starting Small
Test process in small system
C.Elegans (Dalrymple)
Retina (Briggman)
Hippocampal neuroprosthetic (Berger)
Cerebellar neuroprosthetic (Bamford)
Memory from piece of neural tissue (Seung)
Discussion
Good gage of problems – proof of concept!
SI is not new! Many fields can contribute
Tools = problem 1, turning data into model = problem 2
True effort underway – seeking input from SI experts!