SlideShare a Scribd company logo
EXPLORING EMERGENT
PROPERTIES OF
RECURRENT NEURAL
NETWORKS USING A
NOVEL ENERGY FUNCTION
FORMALISM
Rakesh Sengupta
Anindya Pattanayak
Surampudi Bapiraju
INTRODUCTION
 Recurrent neural networks (RNNs) are neural
architectures with feedback loops between
nodes.
 Feedback can originate from the same or
different nodes at each time step, making
RNNs behave like nonlinear dynamical
systems.
 Applications of RNNs include memory
modeling, decision making, and visual sense
of numbers (Sengupta 2014, 2017).
SENGUPTA ET AL 9/25/2023 2
INTRODUCTION
 On-center off-surround recurrent networks
are widely used in various domains, including
short-term memory, decision making, and
pattern recognition.
 These networks are valued for their
versatility and self-organizing capabilities due
to their inherent nonlinear mathematical
properties.
 Traditionally, stability in such networks has
relied on Lyapunov functions or conditions
preventing network divergence.
SENGUPTA ET AL 9/25/2023 3
INTRODUCTION
9/25/2023 SENGUPTA ET AL 4
In this paper, we introduce a novel
method for constructing Lyapunov
functions applicable to recurrent
networks.
We demonstrate the effectiveness
of this approach in specific cases,
highlighting its potential for stability
analysis.
We compare stability criteria
obtained from the energy function
formalism with conventional
ordinary differential equation-based
approaches.
We illustrate how this framework
can be used to make predictions in
real-world biological systems,
drawing from previous research.
ENERGYFUNCTION
FORMALISM
 We begin by considering a single layer
of fully connected recurrent neural
nodes.
 To simplify the equation, we introduce
a variable transformation by letting yi =
xi + Ci
9/25/2023 SENGUPTA ET AL 5
ENERGY
FUNCTION
FORMALISM
9/25/2023 SENGUPTA ET AL 6
The general time evolution of all Cohen-Grossberg
systems can be described by the equation (1)
It is worth noting that this formulation is quite
general and applicable to various network models. It
encompasses additive and shunting model networks,
continuous-time McCulloch-Pitts models, Boltzmann
machines, mean field models, and more.
ENERGYFUNCTION
FORMALISM
 Drawing inspiration from the energy
function principle in classical mechanics,
we can express the relationships
between the derivatives of the
activations xi and the corresponding
energy function Hi as (Eqn 1,2)
 the energy function for a particular
node I can be expressed as (3).
9/25/2023 SENGUPTA ET AL 7
ENERGY
FUNCTION
FORMALISM
 as ሶ
𝑥𝑖 → 0, we have 𝑏𝑖 𝑥𝑖 → ∑𝑐𝑖𝑗 𝑑𝑗 (𝑥𝑗 )
 Thus ignoring the cubic term in the
steady state we can have Eq 3.
9/25/2023 Sengupta et al 8
FULLENERGY
FUNCTION
 Hence, the full energy
function for the system can be
written as (with some changes in
the dummy indices) as Eq 1
 Compare it with Cohen-
Grossberg Liapunov function (eq
2)
SENGUPTA ET AL 9/25/2023 9
APPLICATION:
ADDITIVE
RECURRENT
NETWORK
SENGUPTA ET AL 9/25/2023 10
REACTIONTIMES
(SENGUPTA ET. AL., 2017)
 The driving argument behind the formulation of
reaction time for enumeration was that the reaction
time should correlate with maximum allowed
fluctuation of energy for the network.
9/25/2023 Sengupta et al 11
APPLICATION:
ADDITIVE
RECURRENT
NETWORK
SENGUPTA ET AL 9/25/2023 12
WINNER-
TAKE-ALL
(WTA)
SENGUPTA ET AL 9/25/2023 13
WINNER-
TAKE-ALL
(WTA)
SENGUPTA ET AL 9/25/2023 14
KEYTAKEAWAYS
9/25/2023 Sengupta et al 15
An analytical energy function formalism that effectively derives
the Cohen-Grossberg Lyapunov function for recurrent neural
networks
We compared stability criteria from the energy function
formulation with network analysis and found a close
agreement
For additive networks, the energy-based stability criterion
predicts the onset ofWinner-take-all (WTA) behavior,
complementing the network analysis-based criterion.
We have previously used the energy function to predict
psychophysical attributes, such as reaction times, in human
biological recurrent networks related to the visual sense of
numbers, and these predictions were experimentally verified.
CONTACT:
RAKESH.SENGUPTA@SRU.EDU.IN
Rakesh Sengupta Anindya Pattanayak
Surampudi Bapiraju
9/25/2023 Sengupta et al 16

More Related Content

Similar to RS_Energy_function_V2.pdf

Optimal artificial neural network configurations for hourly solar irradiation...
Optimal artificial neural network configurations for hourly solar irradiation...Optimal artificial neural network configurations for hourly solar irradiation...
Optimal artificial neural network configurations for hourly solar irradiation...
IJECEIAES
 
Energy aware clustering protocol (eacp)
Energy aware clustering protocol (eacp)Energy aware clustering protocol (eacp)
Energy aware clustering protocol (eacp)
IJCNCJournal
 
FLOC: Hesitant Fuzzy Linguistic Term Set Analysis in Energy Harvesting Opport...
FLOC: Hesitant Fuzzy Linguistic Term Set Analysis in Energy Harvesting Opport...FLOC: Hesitant Fuzzy Linguistic Term Set Analysis in Energy Harvesting Opport...
FLOC: Hesitant Fuzzy Linguistic Term Set Analysis in Energy Harvesting Opport...
IJCNCJournal
 
FLOC: HESITANT FUZZY LINGUISTIC TERM SET ANALYSIS IN ENERGY HARVESTING OPPORT...
FLOC: HESITANT FUZZY LINGUISTIC TERM SET ANALYSIS IN ENERGY HARVESTING OPPORT...FLOC: HESITANT FUZZY LINGUISTIC TERM SET ANALYSIS IN ENERGY HARVESTING OPPORT...
FLOC: HESITANT FUZZY LINGUISTIC TERM SET ANALYSIS IN ENERGY HARVESTING OPPORT...
IJCNCJournal
 
IRJET- Sink Mobility based Energy Efficient Routing Protocol for Wireless Sen...
IRJET- Sink Mobility based Energy Efficient Routing Protocol for Wireless Sen...IRJET- Sink Mobility based Energy Efficient Routing Protocol for Wireless Sen...
IRJET- Sink Mobility based Energy Efficient Routing Protocol for Wireless Sen...
IRJET Journal
 
COMPARATIVE STUDY OF BACKPROPAGATION ALGORITHMS IN NEURAL NETWORK BASED IDENT...
COMPARATIVE STUDY OF BACKPROPAGATION ALGORITHMS IN NEURAL NETWORK BASED IDENT...COMPARATIVE STUDY OF BACKPROPAGATION ALGORITHMS IN NEURAL NETWORK BASED IDENT...
COMPARATIVE STUDY OF BACKPROPAGATION ALGORITHMS IN NEURAL NETWORK BASED IDENT...
ijcsit
 
www.ijerd.com
www.ijerd.comwww.ijerd.com
www.ijerd.com
IJERD Editor
 
Iv3515241527
Iv3515241527Iv3515241527
Iv3515241527
IJERA Editor
 
Short Term Load Forecasting: One Week (With & Without Weekend) Using Artifici...
Short Term Load Forecasting: One Week (With & Without Weekend) Using Artifici...Short Term Load Forecasting: One Week (With & Without Weekend) Using Artifici...
Short Term Load Forecasting: One Week (With & Without Weekend) Using Artifici...
IJLT EMAS
 
Throughput analysis of energy aware routing protocol for real time load distr...
Throughput analysis of energy aware routing protocol for real time load distr...Throughput analysis of energy aware routing protocol for real time load distr...
Throughput analysis of energy aware routing protocol for real time load distr...
eSAT Journals
 
Throughput analysis of energy aware routing protocol
Throughput analysis of energy aware routing protocolThroughput analysis of energy aware routing protocol
Throughput analysis of energy aware routing protocol
eSAT Publishing House
 
Computational Investigation of Asymmetric Coplanar Waveguides Using Neural Ne...
Computational Investigation of Asymmetric Coplanar Waveguides Using Neural Ne...Computational Investigation of Asymmetric Coplanar Waveguides Using Neural Ne...
Computational Investigation of Asymmetric Coplanar Waveguides Using Neural Ne...
Konstantinos Karamichalis
 
A Hybrid Deep Neural Network Model For Time Series Forecasting
A Hybrid Deep Neural Network Model For Time Series ForecastingA Hybrid Deep Neural Network Model For Time Series Forecasting
A Hybrid Deep Neural Network Model For Time Series Forecasting
Martha Brown
 
INCREASING WIRELESS SENSOR NETWORKS LIFETIME WITH NEW METHOD
INCREASING WIRELESS SENSOR NETWORKS LIFETIME WITH NEW METHODINCREASING WIRELESS SENSOR NETWORKS LIFETIME WITH NEW METHOD
INCREASING WIRELESS SENSOR NETWORKS LIFETIME WITH NEW METHOD
ijwmn
 
INCREASING WIRELESS SENSOR NETWORKS LIFETIME WITH NEW METHOD
INCREASING WIRELESS SENSOR NETWORKS LIFETIME WITH NEW METHODINCREASING WIRELESS SENSOR NETWORKS LIFETIME WITH NEW METHOD
INCREASING WIRELESS SENSOR NETWORKS LIFETIME WITH NEW METHOD
ijwmn
 
Efficiency of Neural Networks Study in the Design of Trusses
Efficiency of Neural Networks Study in the Design of TrussesEfficiency of Neural Networks Study in the Design of Trusses
Efficiency of Neural Networks Study in the Design of Trusses
IRJET Journal
 
Paper18
Paper18Paper18
QUAD TREE BASED STATIC MULTI HOP LEACH ENERGY EFFICIENT ROUTING PROTOCOL: A N...
QUAD TREE BASED STATIC MULTI HOP LEACH ENERGY EFFICIENT ROUTING PROTOCOL: A N...QUAD TREE BASED STATIC MULTI HOP LEACH ENERGY EFFICIENT ROUTING PROTOCOL: A N...
QUAD TREE BASED STATIC MULTI HOP LEACH ENERGY EFFICIENT ROUTING PROTOCOL: A N...
IJCNCJournal
 
Energy efficient routing algorithm in wireless sensor networks
Energy efficient routing algorithm in wireless sensor networksEnergy efficient routing algorithm in wireless sensor networks
Energy efficient routing algorithm in wireless sensor networks
Alexander Decker
 
Application of nn to power system
Application of nn to power systemApplication of nn to power system
Application of nn to power system
julio shimano
 

Similar to RS_Energy_function_V2.pdf (20)

Optimal artificial neural network configurations for hourly solar irradiation...
Optimal artificial neural network configurations for hourly solar irradiation...Optimal artificial neural network configurations for hourly solar irradiation...
Optimal artificial neural network configurations for hourly solar irradiation...
 
Energy aware clustering protocol (eacp)
Energy aware clustering protocol (eacp)Energy aware clustering protocol (eacp)
Energy aware clustering protocol (eacp)
 
FLOC: Hesitant Fuzzy Linguistic Term Set Analysis in Energy Harvesting Opport...
FLOC: Hesitant Fuzzy Linguistic Term Set Analysis in Energy Harvesting Opport...FLOC: Hesitant Fuzzy Linguistic Term Set Analysis in Energy Harvesting Opport...
FLOC: Hesitant Fuzzy Linguistic Term Set Analysis in Energy Harvesting Opport...
 
FLOC: HESITANT FUZZY LINGUISTIC TERM SET ANALYSIS IN ENERGY HARVESTING OPPORT...
FLOC: HESITANT FUZZY LINGUISTIC TERM SET ANALYSIS IN ENERGY HARVESTING OPPORT...FLOC: HESITANT FUZZY LINGUISTIC TERM SET ANALYSIS IN ENERGY HARVESTING OPPORT...
FLOC: HESITANT FUZZY LINGUISTIC TERM SET ANALYSIS IN ENERGY HARVESTING OPPORT...
 
IRJET- Sink Mobility based Energy Efficient Routing Protocol for Wireless Sen...
IRJET- Sink Mobility based Energy Efficient Routing Protocol for Wireless Sen...IRJET- Sink Mobility based Energy Efficient Routing Protocol for Wireless Sen...
IRJET- Sink Mobility based Energy Efficient Routing Protocol for Wireless Sen...
 
COMPARATIVE STUDY OF BACKPROPAGATION ALGORITHMS IN NEURAL NETWORK BASED IDENT...
COMPARATIVE STUDY OF BACKPROPAGATION ALGORITHMS IN NEURAL NETWORK BASED IDENT...COMPARATIVE STUDY OF BACKPROPAGATION ALGORITHMS IN NEURAL NETWORK BASED IDENT...
COMPARATIVE STUDY OF BACKPROPAGATION ALGORITHMS IN NEURAL NETWORK BASED IDENT...
 
www.ijerd.com
www.ijerd.comwww.ijerd.com
www.ijerd.com
 
Iv3515241527
Iv3515241527Iv3515241527
Iv3515241527
 
Short Term Load Forecasting: One Week (With & Without Weekend) Using Artifici...
Short Term Load Forecasting: One Week (With & Without Weekend) Using Artifici...Short Term Load Forecasting: One Week (With & Without Weekend) Using Artifici...
Short Term Load Forecasting: One Week (With & Without Weekend) Using Artifici...
 
Throughput analysis of energy aware routing protocol for real time load distr...
Throughput analysis of energy aware routing protocol for real time load distr...Throughput analysis of energy aware routing protocol for real time load distr...
Throughput analysis of energy aware routing protocol for real time load distr...
 
Throughput analysis of energy aware routing protocol
Throughput analysis of energy aware routing protocolThroughput analysis of energy aware routing protocol
Throughput analysis of energy aware routing protocol
 
Computational Investigation of Asymmetric Coplanar Waveguides Using Neural Ne...
Computational Investigation of Asymmetric Coplanar Waveguides Using Neural Ne...Computational Investigation of Asymmetric Coplanar Waveguides Using Neural Ne...
Computational Investigation of Asymmetric Coplanar Waveguides Using Neural Ne...
 
A Hybrid Deep Neural Network Model For Time Series Forecasting
A Hybrid Deep Neural Network Model For Time Series ForecastingA Hybrid Deep Neural Network Model For Time Series Forecasting
A Hybrid Deep Neural Network Model For Time Series Forecasting
 
INCREASING WIRELESS SENSOR NETWORKS LIFETIME WITH NEW METHOD
INCREASING WIRELESS SENSOR NETWORKS LIFETIME WITH NEW METHODINCREASING WIRELESS SENSOR NETWORKS LIFETIME WITH NEW METHOD
INCREASING WIRELESS SENSOR NETWORKS LIFETIME WITH NEW METHOD
 
INCREASING WIRELESS SENSOR NETWORKS LIFETIME WITH NEW METHOD
INCREASING WIRELESS SENSOR NETWORKS LIFETIME WITH NEW METHODINCREASING WIRELESS SENSOR NETWORKS LIFETIME WITH NEW METHOD
INCREASING WIRELESS SENSOR NETWORKS LIFETIME WITH NEW METHOD
 
Efficiency of Neural Networks Study in the Design of Trusses
Efficiency of Neural Networks Study in the Design of TrussesEfficiency of Neural Networks Study in the Design of Trusses
Efficiency of Neural Networks Study in the Design of Trusses
 
Paper18
Paper18Paper18
Paper18
 
QUAD TREE BASED STATIC MULTI HOP LEACH ENERGY EFFICIENT ROUTING PROTOCOL: A N...
QUAD TREE BASED STATIC MULTI HOP LEACH ENERGY EFFICIENT ROUTING PROTOCOL: A N...QUAD TREE BASED STATIC MULTI HOP LEACH ENERGY EFFICIENT ROUTING PROTOCOL: A N...
QUAD TREE BASED STATIC MULTI HOP LEACH ENERGY EFFICIENT ROUTING PROTOCOL: A N...
 
Energy efficient routing algorithm in wireless sensor networks
Energy efficient routing algorithm in wireless sensor networksEnergy efficient routing algorithm in wireless sensor networks
Energy efficient routing algorithm in wireless sensor networks
 
Application of nn to power system
Application of nn to power systemApplication of nn to power system
Application of nn to power system
 

Recently uploaded

2001_Book_HumanChromosomes - Genéticapdf
2001_Book_HumanChromosomes - Genéticapdf2001_Book_HumanChromosomes - Genéticapdf
2001_Book_HumanChromosomes - Genéticapdf
lucianamillenium
 
GBSN - Biochemistry (Unit 6) Chemistry of Proteins
GBSN - Biochemistry (Unit 6) Chemistry of ProteinsGBSN - Biochemistry (Unit 6) Chemistry of Proteins
GBSN - Biochemistry (Unit 6) Chemistry of Proteins
Areesha Ahmad
 
Anti-Universe And Emergent Gravity and the Dark Universe
Anti-Universe And Emergent Gravity and the Dark UniverseAnti-Universe And Emergent Gravity and the Dark Universe
Anti-Universe And Emergent Gravity and the Dark Universe
Sérgio Sacani
 
CLASS 12th CHEMISTRY SOLID STATE ppt (Animated)
CLASS 12th CHEMISTRY SOLID STATE ppt (Animated)CLASS 12th CHEMISTRY SOLID STATE ppt (Animated)
CLASS 12th CHEMISTRY SOLID STATE ppt (Animated)
eitps1506
 
Microbiology of Central Nervous System INFECTIONS.pdf
Microbiology of Central Nervous System INFECTIONS.pdfMicrobiology of Central Nervous System INFECTIONS.pdf
Microbiology of Central Nervous System INFECTIONS.pdf
sammy700571
 
TOPIC OF DISCUSSION: CENTRIFUGATION SLIDESHARE.pptx
TOPIC OF DISCUSSION: CENTRIFUGATION SLIDESHARE.pptxTOPIC OF DISCUSSION: CENTRIFUGATION SLIDESHARE.pptx
TOPIC OF DISCUSSION: CENTRIFUGATION SLIDESHARE.pptx
shubhijain836
 
BIOTRANSFORMATION MECHANISM FOR OF STEROID
BIOTRANSFORMATION MECHANISM FOR OF STEROIDBIOTRANSFORMATION MECHANISM FOR OF STEROID
BIOTRANSFORMATION MECHANISM FOR OF STEROID
ShibsekharRoy1
 
JAMES WEBB STUDY THE MASSIVE BLACK HOLE SEEDS
JAMES WEBB STUDY THE MASSIVE BLACK HOLE SEEDSJAMES WEBB STUDY THE MASSIVE BLACK HOLE SEEDS
JAMES WEBB STUDY THE MASSIVE BLACK HOLE SEEDS
Sérgio Sacani
 
Signatures of wave erosion in Titan’s coasts
Signatures of wave erosion in Titan’s coastsSignatures of wave erosion in Titan’s coasts
Signatures of wave erosion in Titan’s coasts
Sérgio Sacani
 
Candidate young stellar objects in the S-cluster: Kinematic analysis of a sub...
Candidate young stellar objects in the S-cluster: Kinematic analysis of a sub...Candidate young stellar objects in the S-cluster: Kinematic analysis of a sub...
Candidate young stellar objects in the S-cluster: Kinematic analysis of a sub...
Sérgio Sacani
 
11.1 Role of physical biological in deterioration of grains.pdf
11.1 Role of physical biological in deterioration of grains.pdf11.1 Role of physical biological in deterioration of grains.pdf
11.1 Role of physical biological in deterioration of grains.pdf
PirithiRaju
 
The cost of acquiring information by natural selection
The cost of acquiring information by natural selectionThe cost of acquiring information by natural selection
The cost of acquiring information by natural selection
Carl Bergstrom
 
Authoring a personal GPT for your research and practice: How we created the Q...
Authoring a personal GPT for your research and practice: How we created the Q...Authoring a personal GPT for your research and practice: How we created the Q...
Authoring a personal GPT for your research and practice: How we created the Q...
Leonel Morgado
 
在线办理(salfor毕业证书)索尔福德大学毕业证毕业完成信一模一样
在线办理(salfor毕业证书)索尔福德大学毕业证毕业完成信一模一样在线办理(salfor毕业证书)索尔福德大学毕业证毕业完成信一模一样
在线办理(salfor毕业证书)索尔福德大学毕业证毕业完成信一模一样
vluwdy49
 
SDSS1335+0728: The awakening of a ∼ 106M⊙ black hole⋆
SDSS1335+0728: The awakening of a ∼ 106M⊙ black hole⋆SDSS1335+0728: The awakening of a ∼ 106M⊙ black hole⋆
SDSS1335+0728: The awakening of a ∼ 106M⊙ black hole⋆
Sérgio Sacani
 
cathode ray oscilloscope and its applications
cathode ray oscilloscope and its applicationscathode ray oscilloscope and its applications
cathode ray oscilloscope and its applications
sandertein
 
Direct Seeded Rice - Climate Smart Agriculture
Direct Seeded Rice - Climate Smart AgricultureDirect Seeded Rice - Climate Smart Agriculture
Direct Seeded Rice - Climate Smart Agriculture
International Food Policy Research Institute- South Asia Office
 
Farming systems analysis: what have we learnt?.pptx
Farming systems analysis: what have we learnt?.pptxFarming systems analysis: what have we learnt?.pptx
Farming systems analysis: what have we learnt?.pptx
Frédéric Baudron
 
Mechanisms and Applications of Antiviral Neutralizing Antibodies - Creative B...
Mechanisms and Applications of Antiviral Neutralizing Antibodies - Creative B...Mechanisms and Applications of Antiviral Neutralizing Antibodies - Creative B...
Mechanisms and Applications of Antiviral Neutralizing Antibodies - Creative B...
Creative-Biolabs
 
Mending Clothing to Support Sustainable Fashion_CIMaR 2024.pdf
Mending Clothing to Support Sustainable Fashion_CIMaR 2024.pdfMending Clothing to Support Sustainable Fashion_CIMaR 2024.pdf
Mending Clothing to Support Sustainable Fashion_CIMaR 2024.pdf
Selcen Ozturkcan
 

Recently uploaded (20)

2001_Book_HumanChromosomes - Genéticapdf
2001_Book_HumanChromosomes - Genéticapdf2001_Book_HumanChromosomes - Genéticapdf
2001_Book_HumanChromosomes - Genéticapdf
 
GBSN - Biochemistry (Unit 6) Chemistry of Proteins
GBSN - Biochemistry (Unit 6) Chemistry of ProteinsGBSN - Biochemistry (Unit 6) Chemistry of Proteins
GBSN - Biochemistry (Unit 6) Chemistry of Proteins
 
Anti-Universe And Emergent Gravity and the Dark Universe
Anti-Universe And Emergent Gravity and the Dark UniverseAnti-Universe And Emergent Gravity and the Dark Universe
Anti-Universe And Emergent Gravity and the Dark Universe
 
CLASS 12th CHEMISTRY SOLID STATE ppt (Animated)
CLASS 12th CHEMISTRY SOLID STATE ppt (Animated)CLASS 12th CHEMISTRY SOLID STATE ppt (Animated)
CLASS 12th CHEMISTRY SOLID STATE ppt (Animated)
 
Microbiology of Central Nervous System INFECTIONS.pdf
Microbiology of Central Nervous System INFECTIONS.pdfMicrobiology of Central Nervous System INFECTIONS.pdf
Microbiology of Central Nervous System INFECTIONS.pdf
 
TOPIC OF DISCUSSION: CENTRIFUGATION SLIDESHARE.pptx
TOPIC OF DISCUSSION: CENTRIFUGATION SLIDESHARE.pptxTOPIC OF DISCUSSION: CENTRIFUGATION SLIDESHARE.pptx
TOPIC OF DISCUSSION: CENTRIFUGATION SLIDESHARE.pptx
 
BIOTRANSFORMATION MECHANISM FOR OF STEROID
BIOTRANSFORMATION MECHANISM FOR OF STEROIDBIOTRANSFORMATION MECHANISM FOR OF STEROID
BIOTRANSFORMATION MECHANISM FOR OF STEROID
 
JAMES WEBB STUDY THE MASSIVE BLACK HOLE SEEDS
JAMES WEBB STUDY THE MASSIVE BLACK HOLE SEEDSJAMES WEBB STUDY THE MASSIVE BLACK HOLE SEEDS
JAMES WEBB STUDY THE MASSIVE BLACK HOLE SEEDS
 
Signatures of wave erosion in Titan’s coasts
Signatures of wave erosion in Titan’s coastsSignatures of wave erosion in Titan’s coasts
Signatures of wave erosion in Titan’s coasts
 
Candidate young stellar objects in the S-cluster: Kinematic analysis of a sub...
Candidate young stellar objects in the S-cluster: Kinematic analysis of a sub...Candidate young stellar objects in the S-cluster: Kinematic analysis of a sub...
Candidate young stellar objects in the S-cluster: Kinematic analysis of a sub...
 
11.1 Role of physical biological in deterioration of grains.pdf
11.1 Role of physical biological in deterioration of grains.pdf11.1 Role of physical biological in deterioration of grains.pdf
11.1 Role of physical biological in deterioration of grains.pdf
 
The cost of acquiring information by natural selection
The cost of acquiring information by natural selectionThe cost of acquiring information by natural selection
The cost of acquiring information by natural selection
 
Authoring a personal GPT for your research and practice: How we created the Q...
Authoring a personal GPT for your research and practice: How we created the Q...Authoring a personal GPT for your research and practice: How we created the Q...
Authoring a personal GPT for your research and practice: How we created the Q...
 
在线办理(salfor毕业证书)索尔福德大学毕业证毕业完成信一模一样
在线办理(salfor毕业证书)索尔福德大学毕业证毕业完成信一模一样在线办理(salfor毕业证书)索尔福德大学毕业证毕业完成信一模一样
在线办理(salfor毕业证书)索尔福德大学毕业证毕业完成信一模一样
 
SDSS1335+0728: The awakening of a ∼ 106M⊙ black hole⋆
SDSS1335+0728: The awakening of a ∼ 106M⊙ black hole⋆SDSS1335+0728: The awakening of a ∼ 106M⊙ black hole⋆
SDSS1335+0728: The awakening of a ∼ 106M⊙ black hole⋆
 
cathode ray oscilloscope and its applications
cathode ray oscilloscope and its applicationscathode ray oscilloscope and its applications
cathode ray oscilloscope and its applications
 
Direct Seeded Rice - Climate Smart Agriculture
Direct Seeded Rice - Climate Smart AgricultureDirect Seeded Rice - Climate Smart Agriculture
Direct Seeded Rice - Climate Smart Agriculture
 
Farming systems analysis: what have we learnt?.pptx
Farming systems analysis: what have we learnt?.pptxFarming systems analysis: what have we learnt?.pptx
Farming systems analysis: what have we learnt?.pptx
 
Mechanisms and Applications of Antiviral Neutralizing Antibodies - Creative B...
Mechanisms and Applications of Antiviral Neutralizing Antibodies - Creative B...Mechanisms and Applications of Antiviral Neutralizing Antibodies - Creative B...
Mechanisms and Applications of Antiviral Neutralizing Antibodies - Creative B...
 
Mending Clothing to Support Sustainable Fashion_CIMaR 2024.pdf
Mending Clothing to Support Sustainable Fashion_CIMaR 2024.pdfMending Clothing to Support Sustainable Fashion_CIMaR 2024.pdf
Mending Clothing to Support Sustainable Fashion_CIMaR 2024.pdf
 

RS_Energy_function_V2.pdf

  • 1. EXPLORING EMERGENT PROPERTIES OF RECURRENT NEURAL NETWORKS USING A NOVEL ENERGY FUNCTION FORMALISM Rakesh Sengupta Anindya Pattanayak Surampudi Bapiraju
  • 2. INTRODUCTION  Recurrent neural networks (RNNs) are neural architectures with feedback loops between nodes.  Feedback can originate from the same or different nodes at each time step, making RNNs behave like nonlinear dynamical systems.  Applications of RNNs include memory modeling, decision making, and visual sense of numbers (Sengupta 2014, 2017). SENGUPTA ET AL 9/25/2023 2
  • 3. INTRODUCTION  On-center off-surround recurrent networks are widely used in various domains, including short-term memory, decision making, and pattern recognition.  These networks are valued for their versatility and self-organizing capabilities due to their inherent nonlinear mathematical properties.  Traditionally, stability in such networks has relied on Lyapunov functions or conditions preventing network divergence. SENGUPTA ET AL 9/25/2023 3
  • 4. INTRODUCTION 9/25/2023 SENGUPTA ET AL 4 In this paper, we introduce a novel method for constructing Lyapunov functions applicable to recurrent networks. We demonstrate the effectiveness of this approach in specific cases, highlighting its potential for stability analysis. We compare stability criteria obtained from the energy function formalism with conventional ordinary differential equation-based approaches. We illustrate how this framework can be used to make predictions in real-world biological systems, drawing from previous research.
  • 5. ENERGYFUNCTION FORMALISM  We begin by considering a single layer of fully connected recurrent neural nodes.  To simplify the equation, we introduce a variable transformation by letting yi = xi + Ci 9/25/2023 SENGUPTA ET AL 5
  • 6. ENERGY FUNCTION FORMALISM 9/25/2023 SENGUPTA ET AL 6 The general time evolution of all Cohen-Grossberg systems can be described by the equation (1) It is worth noting that this formulation is quite general and applicable to various network models. It encompasses additive and shunting model networks, continuous-time McCulloch-Pitts models, Boltzmann machines, mean field models, and more.
  • 7. ENERGYFUNCTION FORMALISM  Drawing inspiration from the energy function principle in classical mechanics, we can express the relationships between the derivatives of the activations xi and the corresponding energy function Hi as (Eqn 1,2)  the energy function for a particular node I can be expressed as (3). 9/25/2023 SENGUPTA ET AL 7
  • 8. ENERGY FUNCTION FORMALISM  as ሶ 𝑥𝑖 → 0, we have 𝑏𝑖 𝑥𝑖 → ∑𝑐𝑖𝑗 𝑑𝑗 (𝑥𝑗 )  Thus ignoring the cubic term in the steady state we can have Eq 3. 9/25/2023 Sengupta et al 8
  • 9. FULLENERGY FUNCTION  Hence, the full energy function for the system can be written as (with some changes in the dummy indices) as Eq 1  Compare it with Cohen- Grossberg Liapunov function (eq 2) SENGUPTA ET AL 9/25/2023 9
  • 11. REACTIONTIMES (SENGUPTA ET. AL., 2017)  The driving argument behind the formulation of reaction time for enumeration was that the reaction time should correlate with maximum allowed fluctuation of energy for the network. 9/25/2023 Sengupta et al 11
  • 15. KEYTAKEAWAYS 9/25/2023 Sengupta et al 15 An analytical energy function formalism that effectively derives the Cohen-Grossberg Lyapunov function for recurrent neural networks We compared stability criteria from the energy function formulation with network analysis and found a close agreement For additive networks, the energy-based stability criterion predicts the onset ofWinner-take-all (WTA) behavior, complementing the network analysis-based criterion. We have previously used the energy function to predict psychophysical attributes, such as reaction times, in human biological recurrent networks related to the visual sense of numbers, and these predictions were experimentally verified.
  • 16. CONTACT: RAKESH.SENGUPTA@SRU.EDU.IN Rakesh Sengupta Anindya Pattanayak Surampudi Bapiraju 9/25/2023 Sengupta et al 16