Recent experimental advances have made it possible to record up to several hundreds of neurons simultaneously in the cortex or in the retina. Analysing such data requires mathematical and numerical methods to describe the spatio-temporal correlations in population activity. This can be done thanks to Maximum Entropy method. Here, a crucial parameter is the product NxR where N is the number of neurons and R the memory depth of correlations (how far in the past does the spike activity affects the current state). Standard statistical mechanics methods are limited to spatial correlation structure with
R = 1 (e.g. Ising model) whereas methods based on transfer matrices, allowing the analysis of spatio-temporal correlations, are limited to NR = 20.
In the first part of the thesis we propose a modified version of the transfer matrix method, based on the parallel version of the Montecarlo algorithm, allowing us to go to NR = 100.
In the second part we present EnaS, a C++ library with a Graphical User Interface developed for neuroscientists. EnaS offers highly interactive tools that allow users to manage data, perform empirical statistics, modeling and visualizing results.
Finally, in a third part, we test our method on synthetic and real data sets. Real data set correspond to retina data provided by neuroscientists partners. Our non extensive analysis shows the advantages of considering spatio-temporal correlations for the analysis of retina spike trains, but it also outlines the limits of Maximum Entropy methods.
For more information about the software that I co-developed with my colleagues, please visit this page:
https://enas.inria.fr/
For more information about the publications, please visit this page:
https://scholar.google.fr/citations?user=L97ZODwAAAAJ
For the thesis, please visit this link:
https://www.theses.fr/178166669
In the first part, Tonda will give a brief introduction to quantum mechanics illustrated on simple discrete systems much like Feynman has in his famous lectures. We will focus on the concepts of spin and photon polarisation.
In the second part we will, one by one, tackle the basics of quantum computation: qubit, quantum computational operation, universal set of quantum instructions and how they relate to Turing machines.
And finally we will take a look at some of the most famous quantum algorithms and for some of them really look under the hood. Also, we will go through a brief overview of the experiments that have been run over the world in past decade or so.
International Conference on Monte Carlo techniques
Closing conference of thematic cycle
Paris July 5-8th 2016
Campus les cordeliers
Jere Koskela's slides
Phase Retrieval: Motivation and TechniquesVaibhav Dixit
Â
This presentation describes two techniques namely Transport of Intensity Equation(TIE) technique and Phase Diversity technique for retrieving phase information from light.
In the first part, Tonda will give a brief introduction to quantum mechanics illustrated on simple discrete systems much like Feynman has in his famous lectures. We will focus on the concepts of spin and photon polarisation.
In the second part we will, one by one, tackle the basics of quantum computation: qubit, quantum computational operation, universal set of quantum instructions and how they relate to Turing machines.
And finally we will take a look at some of the most famous quantum algorithms and for some of them really look under the hood. Also, we will go through a brief overview of the experiments that have been run over the world in past decade or so.
International Conference on Monte Carlo techniques
Closing conference of thematic cycle
Paris July 5-8th 2016
Campus les cordeliers
Jere Koskela's slides
Phase Retrieval: Motivation and TechniquesVaibhav Dixit
Â
This presentation describes two techniques namely Transport of Intensity Equation(TIE) technique and Phase Diversity technique for retrieving phase information from light.
Model Predictive Control based on Reduced-Order ModelsPantelis Sopasakis
Â
The need for reduced-order approximations of dynamical systems emerges naturally in model-based control of very large-scale systems, such as those arising from the discretisation of partial differential equation models. The controller based on the reduced-order model, when in closed-loop with the large-scale system, ought to endow certain properties, in primis stability, but also satisfaction of state constraints and recursive computability of the control law in the case of constrained control.
In this paper we introduce a new approach to the design of model predictive controllers to meet the aforementioned requirements while the on-line complexity is essentially tantamount to the one that corresponds to the low-dimensional approximate model.
Computational Motor Control: Optimal Estimation in Noisy World (JAIST summer ...hirokazutanaka
Â
This is lecure 4 note for JAIST summer school on computational motor control (Hirokazu Tanaka & Hiroyuki Kambara). Lecture video: https://www.youtube.com/watch?v=2-VRBIg5m0w
Computational Motor Control: Optimal Control for Stochastic Systems (JAIST su...hirokazutanaka
Â
This is lecure 5 note for JAIST summer school on computational motor control (Hirokazu Tanaka & Hiroyuki Kambara). Lecture video: https://www.youtube.com/watch?v=XS7MDRMPQfU
In this talk we present a framework for splitting data assimilation problems based upon the model dynamics. This is motivated by assimilation in the unstable subspace (AUS) and center manifold and inertial manifold techniques in dynamical systems. Recent efforts based upon the development of particle filters projected into the unstable subspace will be highlighted.
Computational Motor Control: State Space Models for Motor Adaptation (JAIST s...hirokazutanaka
Â
This is lecure 3 note for JAIST summer school on computational motor control (Hirokazu Tanaka & Hiroyuki Kambara). Lecture video: https://www.youtube.com/watch?v=dtpgJLRt90M
Stochastic Processes describe the system derived by noise.
Level of graduate students in mathematics and engineering.
Probability Theory is a prerequisite.
For comments please contact me at solo.hermelin@gmail.com.
For more presentations on different subjects visit my website at http://www.solohermelin.com.
In this paper, we propose an improved quantum-behaved particle swarm optimization (QPSO), introducing chaos theory into QPSO and exerting logistic map to every particle position X(t) at a certain probability. In this improved QPSO, the logistic map is used to generate a set of chaotic offsets and produce multiple positions around X(t). According to their fitness, the particle's position is updated. In order to further enhance the diversity of particles, mutation operation is introduced into and acts on one dimension of the particle's position. What's more, the chaos and mutation probabilities are carefully selected. Through several typical function experiments, its result shows that the convergence accuracy of the improved QPSO is better than those of QPSO, so it is feasible and effective to introduce chaos theory and mutation operation into QPSO.
Computational Motor Control: Kinematics & Dynamics (JAIST summer course)hirokazutanaka
Â
Computational Motor Control: Kinematics & Dynamics (JAIST summer course)
This is lecture 1 note for JAIST summer school on computational motor control (Hirokazu Tanaka & Hiroyuki Kambara). Lecture video: https://www.youtube.com/watch?v=8nk4DlpAaS8
Future cosmology with CMB lensing and galaxy clusteringMarcel Schmittfull
Â
Next-generation Cosmic Microwave Background experiments such as the Simons Observatory, CMB-S4 and PICO aim to measure gravitational lensing of the Cosmic Microwave Background an order of magnitude better than current experiments. The lensing signal will be highly correlated with measurements of galaxy clustering from next-generation galaxy surveys such as LSST. This will help us understand whether cosmic inflation was driven by a single field or by multiple fields. It will also allow us to accurately measure the growth of structure as a function of time, which is a powerful probe of dark energy and the sum of neutrino masses. I will discuss the prospects for this, as well as recent progress on the theoretical modeling of galaxy clustering, which is key to realize the full potential of these anticipated datasets.
We give an elementary exposition of a method to obtain the infinitesimal point symmetries of Lagrangians.Besides, we exhibit the Lanczos approach to Noetherâs theorem to construct the first integral associated with each symmetry.
MSC 2010:49S05, 58E30, 70H25, 70H33
Model Predictive Control based on Reduced-Order ModelsPantelis Sopasakis
Â
The need for reduced-order approximations of dynamical systems emerges naturally in model-based control of very large-scale systems, such as those arising from the discretisation of partial differential equation models. The controller based on the reduced-order model, when in closed-loop with the large-scale system, ought to endow certain properties, in primis stability, but also satisfaction of state constraints and recursive computability of the control law in the case of constrained control.
In this paper we introduce a new approach to the design of model predictive controllers to meet the aforementioned requirements while the on-line complexity is essentially tantamount to the one that corresponds to the low-dimensional approximate model.
Computational Motor Control: Optimal Estimation in Noisy World (JAIST summer ...hirokazutanaka
Â
This is lecure 4 note for JAIST summer school on computational motor control (Hirokazu Tanaka & Hiroyuki Kambara). Lecture video: https://www.youtube.com/watch?v=2-VRBIg5m0w
Computational Motor Control: Optimal Control for Stochastic Systems (JAIST su...hirokazutanaka
Â
This is lecure 5 note for JAIST summer school on computational motor control (Hirokazu Tanaka & Hiroyuki Kambara). Lecture video: https://www.youtube.com/watch?v=XS7MDRMPQfU
In this talk we present a framework for splitting data assimilation problems based upon the model dynamics. This is motivated by assimilation in the unstable subspace (AUS) and center manifold and inertial manifold techniques in dynamical systems. Recent efforts based upon the development of particle filters projected into the unstable subspace will be highlighted.
Computational Motor Control: State Space Models for Motor Adaptation (JAIST s...hirokazutanaka
Â
This is lecure 3 note for JAIST summer school on computational motor control (Hirokazu Tanaka & Hiroyuki Kambara). Lecture video: https://www.youtube.com/watch?v=dtpgJLRt90M
Stochastic Processes describe the system derived by noise.
Level of graduate students in mathematics and engineering.
Probability Theory is a prerequisite.
For comments please contact me at solo.hermelin@gmail.com.
For more presentations on different subjects visit my website at http://www.solohermelin.com.
In this paper, we propose an improved quantum-behaved particle swarm optimization (QPSO), introducing chaos theory into QPSO and exerting logistic map to every particle position X(t) at a certain probability. In this improved QPSO, the logistic map is used to generate a set of chaotic offsets and produce multiple positions around X(t). According to their fitness, the particle's position is updated. In order to further enhance the diversity of particles, mutation operation is introduced into and acts on one dimension of the particle's position. What's more, the chaos and mutation probabilities are carefully selected. Through several typical function experiments, its result shows that the convergence accuracy of the improved QPSO is better than those of QPSO, so it is feasible and effective to introduce chaos theory and mutation operation into QPSO.
Computational Motor Control: Kinematics & Dynamics (JAIST summer course)hirokazutanaka
Â
Computational Motor Control: Kinematics & Dynamics (JAIST summer course)
This is lecture 1 note for JAIST summer school on computational motor control (Hirokazu Tanaka & Hiroyuki Kambara). Lecture video: https://www.youtube.com/watch?v=8nk4DlpAaS8
Future cosmology with CMB lensing and galaxy clusteringMarcel Schmittfull
Â
Next-generation Cosmic Microwave Background experiments such as the Simons Observatory, CMB-S4 and PICO aim to measure gravitational lensing of the Cosmic Microwave Background an order of magnitude better than current experiments. The lensing signal will be highly correlated with measurements of galaxy clustering from next-generation galaxy surveys such as LSST. This will help us understand whether cosmic inflation was driven by a single field or by multiple fields. It will also allow us to accurately measure the growth of structure as a function of time, which is a powerful probe of dark energy and the sum of neutrino masses. I will discuss the prospects for this, as well as recent progress on the theoretical modeling of galaxy clustering, which is key to realize the full potential of these anticipated datasets.
Similar to Analysis of large scale spiking networks dynamics with spatio-temporal constraints: application to Multi-Electrodes acquisitions in the retina
We give an elementary exposition of a method to obtain the infinitesimal point symmetries of Lagrangians.Besides, we exhibit the Lanczos approach to Noetherâs theorem to construct the first integral associated with each symmetry.
MSC 2010:49S05, 58E30, 70H25, 70H33
Main obstacles of Bayesian statistics or Bayesian machine learning is computing posterior distribution. In many contexts, computing posterior distribution is intractable. Today, there are two main stream to detour directly computing posterior distribution. One is using sampling method(ex. MCMC) and another is Variational inference. Compared to Variational inference, MCMC takes more time and vulnerable to high-dimensional parameters. However, MCMC has strength in simplicity and guarantees of convergence. I'll briefly introduce several methods people using in application.
Universal Approximation Property via Quantum Feature Maps
----
The quantum Hilbert space can be used as a quantum-enhanced feature space in machine learning (ML) via the quantum feature map to encode classical data into quantum states. We prove the ability to approximate any continuous function with optimal approximation rate via quantum ML models in typical quantum feature maps.
---
Contributed talk at Quantum Techniques in Machine Learning 2021, Tokyo, November 8-12 2021.
By Quoc Hoan Tran, Takahiro Goto and Kohei Nakajima
Regularisation & Auxiliary Information in OOD Detectionkirk68
Â
Neural networks are often utilised in critical domain applications (e.g. self-driving cars, financial markets, and aerospace engineering), even though they exhibit overconfident predictions for ambiguous inputs. This deficiency demonstrates a fundamental flaw indicating that neural networks often overfit on spurious correlations. To address this problem in this work we present two objectives that improve the ability of a network to detect out-of-distribution samples and therefore avoid overconfident predictions for ambiguous inputs. We empirically demonstrate that our methods outperform the baseline and performs better than the majority of existing approaches, while still maintaining competitive with the remaining ones. Additionally, we empirically demonstrate the robustness of our approach against common corruptions showcasing the importance of regularisation and auxiliary information in out-of-distribution detection.
Sequence Entropy and the Complexity Sequence Entropy For ððï ActionIJRES Journal
Â
In this paper, we study the complexity of sequence entropy for ðð actions. After that, we define ð¶ðŒ ð¹ðŒ ð , âðŒ ð¹ðŒ ð and the relationships between sequence entropy and complexity sequence entropy. Finally, comparisons between sequence entropy and complexity sequence entropy have been done.
Hierarchical matrix techniques for maximum likelihood covariance estimationAlexander Litvinenko
Â
1. We apply hierarchical matrix techniques (HLIB, hlibpro) to approximate huge covariance matrices. We are able to work with 250K-350K non-regular grid nodes.
2. We maximize a non-linear, non-convex Gaussian log-likelihood function to identify hyper-parameters of covariance.
PROGRAMMA ATTIVITAâ DIDATTICA A.A. 2016/17
DOTTORATO DI RICERCA IN INGEGNERIA STRUTTURALE E GEOTECNICA
____________________________________________________________
STOCHASTIC DYNAMICS AND MONTE CARLO SIMULATION IN EARTHQUAKE ENGINEERING APPLICATIONS
Lecture Series by
Agathoklis Giaralis, Ph.D., M.ASCE., P.E. City, University of London
Visiting Professor Sapienza University of Rome
MVPA with SpaceNet: sparse structured priorsElvis DOHMATOB
Â
The GraphNet (aka S-Lasso), as well as other âsparsity + structureâ priors like TV (Total-Variation), TV-L1, etc., are not easily applicable to brain data because of technical problems
relating to the selection of the regularization parameters. Also, in
their own right, such models lead to challenging high-dimensional optimization problems. In this manuscript, we present some heuristics for speeding up the overall optimization process: (a) Early-stopping, whereby one halts the optimization process when the test score (performance on leftout data) for the internal cross-validation for model-selection stops improving, and (b) univariate feature-screening, whereby irrelevant (non-predictive) voxels are detected and eliminated before the optimization problem is entered, thus reducing the size of the problem. Empirical results with GraphNet on real MRI (Magnetic Resonance Imaging) datasets indicate that these heuristics are a win-win strategy, as they add speed without sacrificing the quality of the predictions. We expect the proposed heuristics to work on other models like TV-L1, etc.
DFA minimization algorithms in map reduceIraj Hedayati
Â
Explaining implementation and analysis of two well known DFA minimisation algorithms namely Morore and Hopcroft, in Map Reduce using Hadoop. Cost analysis and complexity are described.
Please follow this link: http://spectrum.library.concordia.ca/980838/
Similar to Analysis of large scale spiking networks dynamics with spatio-temporal constraints: application to Multi-Electrodes acquisitions in the retina (20)
Show drafts
volume_up
Empowering the Data Analytics Ecosystem: A Laser Focus on Value
The data analytics ecosystem thrives when every component functions at its peak, unlocking the true potential of data. Here's a laser focus on key areas for an empowered ecosystem:
1. Democratize Access, Not Data:
Granular Access Controls: Provide users with self-service tools tailored to their specific needs, preventing data overload and misuse.
Data Catalogs: Implement robust data catalogs for easy discovery and understanding of available data sources.
2. Foster Collaboration with Clear Roles:
Data Mesh Architecture: Break down data silos by creating a distributed data ownership model with clear ownership and responsibilities.
Collaborative Workspaces: Utilize interactive platforms where data scientists, analysts, and domain experts can work seamlessly together.
3. Leverage Advanced Analytics Strategically:
AI-powered Automation: Automate repetitive tasks like data cleaning and feature engineering, freeing up data talent for higher-level analysis.
Right-Tool Selection: Strategically choose the most effective advanced analytics techniques (e.g., AI, ML) based on specific business problems.
4. Prioritize Data Quality with Automation:
Automated Data Validation: Implement automated data quality checks to identify and rectify errors at the source, minimizing downstream issues.
Data Lineage Tracking: Track the flow of data throughout the ecosystem, ensuring transparency and facilitating root cause analysis for errors.
5. Cultivate a Data-Driven Mindset:
Metrics-Driven Performance Management: Align KPIs and performance metrics with data-driven insights to ensure actionable decision making.
Data Storytelling Workshops: Equip stakeholders with the skills to translate complex data findings into compelling narratives that drive action.
Benefits of a Precise Ecosystem:
Sharpened Focus: Precise access and clear roles ensure everyone works with the most relevant data, maximizing efficiency.
Actionable Insights: Strategic analytics and automated quality checks lead to more reliable and actionable data insights.
Continuous Improvement: Data-driven performance management fosters a culture of learning and continuous improvement.
Sustainable Growth: Empowered by data, organizations can make informed decisions to drive sustainable growth and innovation.
By focusing on these precise actions, organizations can create an empowered data analytics ecosystem that delivers real value by driving data-driven decisions and maximizing the return on their data investment.
Data Centers - Striving Within A Narrow Range - Research Report - MCG - May 2...pchutichetpong
Â
M Capital Group (âMCGâ) expects to see demand and the changing evolution of supply, facilitated through institutional investment rotation out of offices and into work from home (âWFHâ), while the ever-expanding need for data storage as global internet usage expands, with experts predicting 5.3 billion users by 2023. These market factors will be underpinned by technological changes, such as progressing cloud services and edge sites, allowing the industry to see strong expected annual growth of 13% over the next 4 years.
Whilst competitive headwinds remain, represented through the recent second bankruptcy filing of Sungard, which blames âCOVID-19 and other macroeconomic trends including delayed customer spending decisions, insourcing and reductions in IT spending, energy inflation and reduction in demand for certain servicesâ, the industry has seen key adjustments, where MCG believes that engineering cost management and technological innovation will be paramount to success.
MCG reports that the more favorable market conditions expected over the next few years, helped by the winding down of pandemic restrictions and a hybrid working environment will be driving market momentum forward. The continuous injection of capital by alternative investment firms, as well as the growing infrastructural investment from cloud service providers and social media companies, whose revenues are expected to grow over 3.6x larger by value in 2026, will likely help propel center provision and innovation. These factors paint a promising picture for the industry players that offset rising input costs and adapt to new technologies.
According to M Capital Group: âSpecifically, the long-term cost-saving opportunities available from the rise of remote managing will likely aid value growth for the industry. Through margin optimization and further availability of capital for reinvestment, strong players will maintain their competitive foothold, while weaker players exit the market to balance supply and demand.â
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...Subhajit Sahu
Â
Abstract â Levelwise PageRank is an alternative method of PageRank computation which decomposes the input graph into a directed acyclic block-graph of strongly connected components, and processes them in topological order, one level at a time. This enables calculation for ranks in a distributed fashion without per-iteration communication, unlike the standard method where all vertices are processed in each iteration. It however comes with a precondition of the absence of dead ends in the input graph. Here, the native non-distributed performance of Levelwise PageRank was compared against Monolithic PageRank on a CPU as well as a GPU. To ensure a fair comparison, Monolithic PageRank was also performed on a graph where vertices were split by components. Results indicate that Levelwise PageRank is about as fast as Monolithic PageRank on the CPU, but quite a bit slower on the GPU. Slowdown on the GPU is likely caused by a large submission of small workloads, and expected to be non-issue when the computation is performed on massive graphs.
Explore our comprehensive data analysis project presentation on predicting product ad campaign performance. Learn how data-driven insights can optimize your marketing strategies and enhance campaign effectiveness. Perfect for professionals and students looking to understand the power of data analysis in advertising. for more details visit: https://bostoninstituteofanalytics.org/data-science-and-artificial-intelligence/
Chatty Kathy - UNC Bootcamp Final Project Presentation - Final Version - 5.23...John Andrews
Â
SlideShare Description for "Chatty Kathy - UNC Bootcamp Final Project Presentation"
Title: Chatty Kathy: Enhancing Physical Activity Among Older Adults
Description:
Discover how Chatty Kathy, an innovative project developed at the UNC Bootcamp, aims to tackle the challenge of low physical activity among older adults. Our AI-driven solution uses peer interaction to boost and sustain exercise levels, significantly improving health outcomes. This presentation covers our problem statement, the rationale behind Chatty Kathy, synthetic data and persona creation, model performance metrics, a visual demonstration of the project, and potential future developments. Join us for an insightful Q&A session to explore the potential of this groundbreaking project.
Project Team: Jay Requarth, Jana Avery, John Andrews, Dr. Dick Davis II, Nee Buntoum, Nam Yeongjin & Mat Nicholas
As Europe's leading economic powerhouse and the fourth-largest hashtag#economy globally, Germany stands at the forefront of innovation and industrial might. Renowned for its precision engineering and high-tech sectors, Germany's economic structure is heavily supported by a robust service industry, accounting for approximately 68% of its GDP. This economic clout and strategic geopolitical stance position Germany as a focal point in the global cyber threat landscape.
In the face of escalating global tensions, particularly those emanating from geopolitical disputes with nations like hashtag#Russia and hashtag#China, hashtag#Germany has witnessed a significant uptick in targeted cyber operations. Our analysis indicates a marked increase in hashtag#cyberattack sophistication aimed at critical infrastructure and key industrial sectors. These attacks range from ransomware campaigns to hashtag#AdvancedPersistentThreats (hashtag#APTs), threatening national security and business integrity.
ð Key findings include:
ð Increased frequency and complexity of cyber threats.
ð Escalation of state-sponsored and criminally motivated cyber operations.
ð Active dark web exchanges of malicious tools and tactics.
Our comprehensive report delves into these challenges, using a blend of open-source and proprietary data collection techniques. By monitoring activity on critical networks and analyzing attack patterns, our team provides a detailed overview of the threats facing German entities.
This report aims to equip stakeholders across public and private sectors with the knowledge to enhance their defensive strategies, reduce exposure to cyber risks, and reinforce Germany's resilience against cyber threats.
6. Probabilistic
Models
Maximum
entropy
1 time-step memory
(Marre et al 09)
Generalized
Linear model
Point
process
General framework
(Vasquez et al 12)
Ising
(Schneidman
et al 06)
Triplets (Ganmor et al
09)
Spatio-
Temporal
Spatial
No memory
Limited to
1 time step
memory
Limited to
small scale
Neurons are considered
conditionally independent
given the past
Hawks
Linear Non
Linear model
# of neurons doubles
every 8 years !!
6
7. Goal
⢠Definitions
â Basic concepts
â Maximum entropy principle (Spatial
& Spatio-temporal).
⢠Montecarlo in the service of
large neural spike trains
⢠Fitting parameters
â Tests on synthetic data
â Application on real data
⢠The EnaS software
⢠Discussion
Develop a
framework to fit
spatio temporal
maximum entropy
models on large
scale spike trains
7
8. Goal
⢠Definitions
â Basic concepts
â Maximum entropy principle (Spatial
& Spatio-temporal).
⢠Montecarlo in the service of
large neural spike trains
⢠Fitting parameters
â Tests on synthetic data
â Application on real data
⢠The EnaS software
⢠Discussion
Develop a
framework to fit
spatio temporal
maximum entropy
models on large
scale spike trains
8
23. â ð€â² ð€ = ðâ ð0
ð· , if ð€â² ð€ is a legal transition
0 Otherwise
Transfer matrix
Non normalized
Perron-Frobenius
Theorem
Right eigenvector
Left eigenvector
ð ð
The biggest eigenvalue
ð (. )
L(. )
Using Chapmanâ
Kolmogorov equation
ð ð0
ð
=
ðâ ð0
ð
ð ð
ðâð·+1 ð ð ðâð·
ð
ð¿(ð0
ð·
)
ð« â = log ð ð
Direct computing of
the Kullback-Leibler
Divergence
ð ðŸð¿ ð ð
ð
, ð ð = ð« â â ð ð
ð
â â ð® ð ð
ð
2 ðð·
Compute the
average of
monomials
ð ðð =
ðð« â
ððð
23
Pressure Entropy
Empirical
probability
of the
potential
24. Setting the
constraints
Computing the
empirical
distribution
ð ð
ð
(ðð)
Random set of
parameters
Computing the
predicted
distribution
ð ð(ðð¡) ðð
Update the
parameters
Final set of
parameters
Predicted
distribution
Comparison
Transfer
Matrix
24
25. Limitation of the transfer matrix â ð€ð€
â²
2 ðð·
2 ðð·
â ð€â² ð€ ð, ð â â ð·ðð¢ððð = 1 ðµðŠð¡ð
2 ðð· à 2 ðð· = 22ðð· ðµðŠð¡ðð â ð€â² ð€
Memoryneed
Neuron number
Range: R = D+1 = 3
20 neurons
ïš
1,099,511,627,776 ððµ
25
26. ðð = 20
Small scale Large scale
ðð > 20ðð †20
Transfer matrix Montecarlo
26
27. Computing the
predicted
distribution
ð ð(ðð¡) ðð
Setting the
constraints
Computing the
empirical
distribution
ð ð
ð
(ðð)
Random set of
parameters
Update the
parameters
Final set of
parameters
Predicted
distribution
Comparison
Transfer
Matrix
27
28. Goal
Develop a
framework to fit
spatio temporal
maximum entropy
models on large
scale spike trains
⢠Definitions
â Basic concepts
â Maximum entropy principle (Spatial
& Spatio-temporal).
⢠Montecarlo in the service of
large neural spike trains
⢠Fitting parameters
â Tests on synthetic data
â Application on real data
⢠The EnaS software
⢠Discussion
28
34. Algorithm review Start: Random spike
train
Parameters ð
ð neurons.
Length = ð
Choose a random
event and flip it
Compute ðÎâ ð
ðÎâ ð > ð
ð â [0,1]
Choose between
[ð· + 1, ð â ð· â 1]
No
Accept the
change
Reject the
change
Yes
Updated Montecarlo
spike train
ððððð / Loop
34
Computed only
between [âð·, +ð·]
35. Hassan Nasser, Olivier Marre, and Bruno Cessac. Spike trains analysis using
Gibbs distributions and Montecarlo method. Journal of Statistical Mechanics:
Theory and experiments, 2013.
35
37. Goal
Develop a
framework to fit
spatio temporal
maximum entropy
models on large
scale spike trains
⢠Definitions
â Basic concepts
â Maximum entropy principle (Spatial
& Spatio-temporal).
⢠Montecarlo in the service of
large neural spike trains
⢠Fitting parameters
â Tests on synthetic data
â Application on real data
⢠The EnaS software
⢠Discussion
37
38. Fitting parameters / concept
Maximizing entropy
(difficult because computing the exact entropy intractable)
â¡
minimizing the divergence
ð ðŸð¿ ð ð
ð
, ð ð = ð« ð â ð ð
ð
â â ð®[ð ð
ð
]
DudÃk, M., Phillips, S., and Schapire, R. (2004). Performance guarantees for
regularized maximum entropy density estimation. Proceedings of the 17th Annual
Conference on Computational Learning Theory.
Small scale: easy to compute
Large scale: hard to compute
- Bounding the negative log likelihood
Divergence
Iterations
Big ð ðŸð¿
Small ð ðŸð¿
38
- Relaxation
39. Fitting parameters / concept
Maximizing entropy
(difficult because computing the exact entropy intractable)
â¡
minimizing the divergence
ð ðŸð¿ ð ð
ð
, ð ð = ð« ð â ð ð
ð
â â ð®[ð ð
ð
]
DudÃk, M., Phillips, S., and Schapire, R. (2004). Performance guarantees for
regularized maximum entropy density estimation. Proceedings of the 17th Annual
Conference on Computational Learning Theory.
Small scale: easy to compute
Large scale: hard to compute
- Bounding the negative log likelihood
Divergence
Iterations
Big ð ðŸð¿
Small ð ðŸð¿
39
- Relaxation
40. Fitting parameters / concept
Hassan Nasser and Bruno Cessac. Parameters fitting for spatio-temporal
maximum entropy distributions: application to neural spike trains. Submitted to
Entropy.
- Bounding the Divergence
40
- With relaxation
42. Setting the
constraints
Computing the
empirical
distribution
ð ð
ð
(ðð)
Random set of
parameters
Computing the
predicted
distribution
ð ð(ðð¡) ðð
Update the
parameters
Final set of
parameters
Exact predicted
distribution
Comparison
Monte-
Carlo
Fitting
42
43. Updating the target distribution
ð2
ð«[ð ]
ððð ðð ð
=
ð=ââ
â
ð¶ðð ð
ð ð+ð¹ ðð = ð ð ðð +
ð
ð2
ð«[ð ]
ððð ðð ð
ð¿ ð +
1
2
ð,ð,ð
ð3
ð«[ð]
ððð ðð ð ððð
ð¿ð ð¿ ð ð¿ð + â¯
New ð ïš New distribution.
Montecarlo
Taylor
Expansion
Previous
distribution
Exponential decay of correlation
ïš
In practice n is finite
(If ð¹ ððððð)ð = ðð¢ððð¡ððð(ð)
43
âððð£ðŠ ð¡ð ððððð¢ð¡ð
50. Data Courtesy: Michael J. Berry II (Princeton University)
and Olivier Marre (Institut de la vision, Paris).
Purely spatial pairwise
Pairwise with 1 time-step memory
Binned at 20 ms
Application on retinal data
50
Schneidman et al 2006
51. Real data: 20 neurons
Spatial Pairwise Spatio-temporal Pairwise
51
52. Real data: 40 neurons
Pairwise Spatial Pairwise Spatio-temporal
52
53. Goal
Develop a
framework to fit
spatio temporal
maximum entropy
models on large
scale spike trains
⢠Definitions
â Basic concepts
â Maximum entropy principle (Spatial
& Spatio-temporal).
⢠Montecarlo in the service of
large neural spike trains
⢠Fitting parameters
â Tests on synthetic data
â Application on real data
⢠The EnaS software
⢠Discussion
53
61. Map: C++ data container Sorting in a chosen order
âŠ
61
62. Map: C++ data container Sorting in a chosen order
âŠ
It appeared two times!
62
63. Architecture
EnaS
RasterBlock Gibbs Potential
Graphical User
interface
- Data management
- Grammar
- Empirical statistics
- âŠ
- Defining models
- Generating artificial
spike trains
- Fitting
- Montecarlo process
(Parallelization)
- Interactive environment
- Visualization of
stimulus and response
simultaneously.
- Demo
63
64. Parallelization of Montecarlo process
ðð¡ðððð
xx
x
x x
x
x x
x
x x
x x
x
x x
x
x
x x
Personal Multi-processors computer: 2-8 processors
Cluster (64 processors machines at INRIA)
OpenMp
64
MPI More processors / More time consuming in our case
72. Synthetic data Vs Real data
Synthetic data
Potential shape is known
(monomials are known)
Real data
72
Potential shape is
unknown
(monomials are
unknown)Fitting only
Guessing the shape
+
Fitting
73. Monomials
Model
Canonical
Ising, pairwise with
delay, triplets, âŠ
Small scaleLarge scale
- Big computation time
- Non Observed
monomials
- Estimation errors.
Pre-Selection
73
Rodrigo Cofre & Bruno Cessac
40 neurons
74. Making sense of parameters
Model
parameters
Evaluate the importance
of particular type of
correlations
Possibility of generalize
the model prediction on
new stimulus
74
76. EnaS
Retina Spike sorting Spike trainStimulus
Visualization
Visualization
+
Empirical analysis
+
Maximum Entropy
modelling
NowFuture
- More empirical
observation
packages
- More neural
coding
functionalities
Spike sorting
- Receptive
field
- Neurons
selection
Type
identification
- Stimulus
design
- Features
extraction
76
Retina
models
VirtualRetina
77. Next âŠ
Starting a company in IT/Data Analytics:
â First prize in innovative project competition (UNICE Foundation).
â Current project: Orientation in education using real surveys.
â EnaS is in perspective in collaboration with INRIA.
Caty Conraux & Vincent Tricard
77
80. Appendix
⢠Tuning Ntimes.
⢠Tuning Nflip.
⢠Validating montecarlo algorithm.
⢠Tunnig delta.
⢠MPI Vs OpenMP, memory.
⢠Why MPI is not better than OpenMP?
⢠Computational complexity of the Montecarlo algorithm.
⢠Review of Montecarlo / Nflip.
⢠Number of Iterations for fitting.
⢠Fluctuations on parameters / Non existing monomials.
⢠Epsilon on fitting parameters.
⢠Binning.
⢠Tests with several stimulus.
⢠Granot-Atedgi et al 2013
⢠Granot-Atedgi et al 2013
80
85. ⢠Multiprocessors computers:
â Personal computer (2-8 processors).
â Cluster (64 processors machines at INRIA).
⢠Parallel programming frameworks:
â OpenMp: The processors of the same computer divide
the tasks (live memory (RAM) is shared).
â MPI: several processors on each computer share the
task (Memory in not shared).
4 processors
ïš Time/4.
Parallelization
64 processors ïš Time/64.
85
86. MPI
⢠OpenMP is limited to the number of processors
on a single machine.
⢠With MPI, 64 processors x 10 machine ïš 640
processors.
⢠Although we though it would take less time
with MPI, but âŠ! Master
computer1 cluster of 64 proc
Another cluster of 64 proc
Another cluster of 64 proc
Another cluster of 64 proc
The whole
Montecarlo
Spike train
At each change of the memory, there will be a communication between the clusters and the master
ïšAt each flip ïš loss of time in communication more than computing 86
88. Computational complexity
Taken for running this algorithm:
ð¶ðððð¢ð¡ððð ð¡ððð = ð. ð. ðð¡ðððð . ð¡Îâ ð
ð¡Îâ ð
= ðð¡(ð¿)
Start: Random
spike train
Choose a random
event and flip it
Compute ðÎâ ð
ðÎâ ð > ð
ð â [0,1]
No
Accept the
change
Reject the
change
Yes
Updated Montecarlo
spike train
Loop : ð. ð. ðð¡ðððð 2- In each loop, computing ðÎâ ð needs to
perform a loop over the monomials.
1- We have a loop over ð. ð. ðð¡ðððð .
On a cluster of 64 processors:
- 40 Neurons Ising: 10 min
- 40 Neurons Pairwise: 20 min
88
89. Start: Random spike
train
Parameters ð
ð neurons.
Length = ðð¡ðððð
Choose a random
event and flip it
Compute ðÎâ ð
ðÎâ ð > ð
ð â [0,1]
No
Accept the
change
Reject the
change
Yes
Updated Montecarlo
spike train
Tuning
Loop : ððððð=
Algorithm
review
ð Ã ð Ã ðð¡ðððð
Choose between
[ð·, ðð¡ðððð â ð·]
89
90. How many iterations do we need?
⢠ðð < 20:
â 50 parallel + 100 sequential
⢠ðð < 150:
â 100 parallel + 100
sequential
90
91. ð on parameters fitting
⢠Dudik et al does not allow that:
⢠ðœð > ðð || ðœð > 1 â ðð. In this case â ðœð = 0.9 ðð
⢠We avoid dividing by 0 (
âŠ
ð ð
) ⊠by replacing
putting ðð = ââ
91
92. Problem of non- observed monomials
Central limit theorem (Fluctuations on monomials averages)
ïšð ð ðð =
ðð«
ðð
; ð ð
â
ðð + ðŒ =
ðð«
ðð
+
ðð2 ð«
ðð2 + ⊠=
ðð«
ðð
+ ðð³ + â¯
ïšð = ðâ + ð : Fluctuations on parameters
ïšð = ð³â1 ðŒ
Covariance matrix ð³ðð =
ð2 ð«
ðð ð ðð ð
Convex
Computing ð³ over 1000 potential
shows that a big percentage of ð³ is
zero ïš ð³â1
will have big value ïš
flucutations on ð are big.
ð³
92
93. Binning
⢠Change completely
the statistics.
⢠700% of more new
patterns appear when
we bin at 20.
⢠Should be studied
rigorously.
93
95. 95
- Loss of information
- Loosing biological scale
- More dense spike train
- Less non-observed monomials
Why spike trains have been binned in the literature?
- No clear answer.
- Relation between taking binning as a substitute for
memory is not convincing.
- Might be because it allows having more monomials ïš
Less dangerous for convexity ïš convergence is more
guaranteed.
96. Making sense of parameters
Stimulus 1
Stimulus 2
Stimulus 4
= ð ð ð1
= ð ð ð2
= ð ð ð3
= ð ð ð4
Stimulus 3
96