Oscillatory Neural Network for Image Segmentation with Biased Competition for Attention

Raiko, Tapani; Valpola, Harri

doi:10.1007/978-1-4614-0164-3_7

Cited by 6 publications

(3 citation statements)

References 15 publications

(13 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Two interacting two-dimensional oscillator networks with different couplings within rows and within columns are used. [48,170] • Oscillator network-based Ising machine for solving combinatorial optimization problems. [49][50][51] A two-dimensional network of interacting oscillators is the central feature of almost all of the proposed neuromorphic approaches to computing.…”

Section: Neuromorphic Computing Using 2d Shno Arraysmentioning

confidence: 99%

Current Modulation of Nanoconstriction Spin-Hall Nano-Oscillators

et al. 2017

View full text Add to dashboard Cite

This thesis explores the vibrant phenomena of spin Hall nano-oscillators (SHNOs); from wideband oscillation at the GHz range, through propagating spin-wave emission, to mutual synchronization in two-dimensional SHNO arrays, and tries to lay the foundation for a far-reaching SHNO technology, targeting magnonics, microwave signal generation, and neuromorphic computing. After a short introduction to the theoretical background in Chapter 1, in Chapter 2, ways of improving the spin transport properties between NiFe and Pt are explored: an 0.4 nm ultra-thin layer of Hf at the NiFe/Pt interface is found to reduce the threshold current by 20% as a result of the change in spin mixing conductance G Ö eff. Then, W/CoFeB/MgO stacks with perpendicular magnetic anisotropy (PMA) are used to demonstrate wide frequency tunability and sub-mA threshold currents in CMOS compatible SHNOs. By further increasing the PMA, the auto-oscillation frequency exceeds the ferromagnetic resonance (FMR) frequency, turning the SHNO into a propagating spin-wave emitter in which the propagation wave-vector is tunable with applied current and field. Finally, a GHz nano-scale SHNO modulated by an 80 MHz radio-frequency (RF) current is presented. The modulation needs no bulky microwave mixer, promising a compact modulator unit. Chapter 3 introduces two-dimensional SHNO arrays. Robust mutual synchronization is demonstrated in arrays accommodating up to 64 oscillators, achieving record high quality factors of 170,000 at an operating frequency of 10 GHz. Injection of two external microwave signals reproduces the two-dimensional synchronization maps used in neuromorphic vowel recognition. Chapter 4 emphasizes the importance of individual SHNO control in arrays. Gated SHNOs are demonstrated with substantial voltage tuning of the threshold current and the SHNO frequency. Voltage controlled mutual synchronization is also demonstrated. The MgO/AlO x /Si 3 N 4 gate is found to exhibit a memristive behavior governed by ion migration and acts as an embedded memory making each SHNO a complete non-volatile oscillator with integrated weights. The exciting nature of coupled non-volatile oscillator arrays controlled by electric field could lead to a paradigm shift in non-conventional computing. Chapter 5 discusses the implications of the demonstrated SHNO technology and how it may impact future applications.

show abstract

Section: Neuromorphic Computing Using 2d Shno Arraysmentioning

confidence: 99%

Current Modulation of Nanoconstriction Spin-Hall Nano-Oscillators

et al. 2017

View full text Add to dashboard Cite

show abstract

“…Oscillations are in a focus of neuroscience, since they correlate with many cognitive tasks. For example, oscillatory neural networks are productive for investigation of image recognition [13,14], as well as activating network states associated with memory recall [15]. They became the core of interdisciplinary research which unites psychophysics, neuroscience, cognitive psychology, biophysics, and computational modeling [16,17].…”

Section: Introductionmentioning

confidence: 99%

Inertial Neural Networks with Unpredictable Oscillations

2020

View full text Add to dashboard Cite

In this paper, inertial neural networks are under investigation, that is, the second order differential equations. The recently introduced new type of motions, unpredictable oscillations, are considered for the models. The motions continue a line of periodic and almost periodic oscillations. The research is of very strong importance for neuroscience, since the existence of unpredictable solutions proves Poincaré chaos. Sufficient conditions have been determined for the existence, uniqueness, and exponential stability of unpredictable solutions. The results can significantly extend the role of oscillations for artificial neural networks exploitation, since they provide strong new theoretical and practical opportunities for implementation of methods of chaos extension, synchronization, stabilization, and control of periodic motions in various types of neural networks. Numerical simulations are presented to demonstrate the validity of the theoretical results.

show abstract

“…The investigation of neural networks with chaotic oscillations has been of strong interest of numerous researchers [13][14][15]. Chaos is useful in many applications of neural networks such as separating image segments [16][17][18][19], synchronization [20][21][22], pattern recognition [23] and information processing [24].…”

Section: Introductionmentioning

confidence: 99%

Strongly Unpredictable Oscillations of Hopfield-Type Neural Networks

2020

View full text Add to dashboard Cite

In this paper, unpredictable oscillations in Hopfield-type neural networks is under investigation. The motion strongly relates to Poincaré chaos. Thus, the importance of the dynamics is indisputable for those problems of artificial intelligence, brain activity and robotics, which rely on chaos. Sufficient conditions for the existence and uniqueness of exponentially stable unpredictable solutions are determined. The oscillations continue the line of periodic and almost periodic motions, which already are verified as effective instruments of analysis and applications for image recognition, information processing and other areas of neuroscience. The concept of strongly unpredictable oscillations is a significant novelty of the present research, since the presence of chaos in each coordinate of the space state provides new opportunities in applications. Additionally to the theoretical analysis, we have provided strong simulation arguments, considering that all of the assumed conditions are fulfilled.

show abstract

Oscillatory Neural Network for Image Segmentation with Biased Competition for Attention

Cited by 6 publications

References 15 publications

Current Modulation of Nanoconstriction Spin-Hall Nano-Oscillators

Current Modulation of Nanoconstriction Spin-Hall Nano-Oscillators

Inertial Neural Networks with Unpredictable Oscillations

Strongly Unpredictable Oscillations of Hopfield-Type Neural Networks

Contact Info

Product

Resources

About