A fast fixed point learning method to implement associative memory on CNNs

Szolgay, Péter; Szatmári, István; Laszlo, K.

doi:10.1109/81.563627

Cited by 29 publications

(17 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Space-variant linear programming is required by number of recently published CNN solutions (e.g. Reference [54] and see also the non-linear non-PDE-related di usion model in this paper).…”

Section: Implementation Of the Algorithm In A Cnn Universal Chip Envimentioning

confidence: 99%

See 1 more Smart Citation

CNN architectures for constrained diffusion based locally adaptive image processing

Rekeczky¹

2002

Circuit Theory & Apps

View full text Add to dashboard Cite

SUMMARYIn this paper, a cellular neural network (CNN) based locally adaptive scheme is presented for image segmentation and edge detection. It is shown that combining a constrained (linear or non-linear) di usion approach with adaptive morphology leads to a robust segmentation algorithm for an important class of image models. These images are comprised of simple geometrical objects, each having a homogeneous grey-scale level and they might be overlapping. The background illumination is inhomogeneous, the objects are corrupted by additive Gaussian noise and possibly blurred by low-pass-ÿltering-type e ects. Typically, this class has a multimodal (in most cases bimodal) image histogram and no special (easily exploitable) characteristics in the frequency domain. The synthesized analogic (analog and logic) CNN algorithm combines a di usion-type ÿltering with a locally adaptive strategy based on estimating the ÿrst-order (mean) and second-order (variance) statistics. Both PDE-and non-PDE-related di usion schemes are examined and compared in the CNN framework. It is shown that the proposed algorithm with various di usion-type ÿlters o ers a more robust solution than some globally optimal thresholding schemes. All algorithmic steps are realized using nearest-neighbour CNN templates. The VLSI implementation complexity and some robustness issues are carefully analysed and discussed in detail. A number of tests have been completed on original and artiÿcial grey-scale images.

show abstract

“…Space-variant linear programming is required by number of recently published CNN solutions (e.g. Reference [54] and see also the non-linear non-PDE-related di usion model in this paper).…”

Section: Implementation Of the Algorithm In A Cnn Universal Chip Envimentioning

confidence: 99%

“…and software packages (e.g. Reference [54][55][56]) have been developed and are available for optimizing similar templates both for software simulators and for the currently available CNN Universal Chips (References [5][6][7][8][9][10]). …”

Section: Stability and Robustness Of Linear And Non-linear Templatesmentioning

confidence: 99%

CNN architectures for constrained diffusion based locally adaptive image processing

Rekeczky¹

2002

Circuit Theory & Apps

View full text Add to dashboard Cite

show abstract

“…1 the authors use Aii.ij as a design parameter to guarantee the storage of the desired pattems. However, increasing the self-feedback entails an increment in the number of spurious states, and then a reduction of error correction capability [7]. So a trade-off must be pursued between these two conflicting aspects.…”

Section: Experiments and Resultsmentioning

confidence: 99%

“…As concerns this last point, we note that if is allowed to increase, inequalities (4) always have a solution, for every value of 6. In[7] …”

mentioning

confidence: 98%

“…As a consequence of this last variant, convergence cannot be proved as in[2,5].In [7] the authors propose the same method to design CNNs with analog weights. The present approach differ from[7] in two aspects: the digital representation of weights, and the value of Aij,ij , here fixed at one. As concerns this last point, we note that if is allowed to increase, inequalities (4) always have a solution, for every value of 6.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Hardware-oriented algorithm for associative memories on cellular neural networks

Perfetti¹,

Salerno²,

Costantini³

Proceedings of the 2002 7th IEEE International Workshop on Cellular Neural Networks and Their Applications

View full text Add to dashboard Cite

1. 00133 Rome, Irdy -emoil: cosranlini@uniromo2.irIn this paper, we present a new laming algorithm used to implement associative memories an digital cellular neural networks. The algorithm can be easily implemented in hardware or simulated on a digid compute without numerical errom. These attractive f e a m came fmm the finite precisian of connection weights, automatically taken into account as a design constraint; moreover. no multiplication is needed for weight computation.

show abstract

Associative Memory Networks

Swamy²

2013

Neural Networks and Statistical Learning

View full text Add to dashboard Cite

A fast fixed point learning method to implement associative memory on CNNs

Cited by 29 publications

References 8 publications

CNN architectures for constrained diffusion based locally adaptive image processing

CNN architectures for constrained diffusion based locally adaptive image processing

Hardware-oriented algorithm for associative memories on cellular neural networks

Associative Memory Networks

Contact Info

Product

Resources

About