Go is a large array. Very large arrays of neural networks with weighting factors and iteration in formulas that can be dynamically modified programs using FPGA’s.
The Chinese announced a optical quantum computer in physical.org on 12/10/20, as reported by NHK TV. It was ~ 100 billion times more powerful than MIT/Google’s project, which used to be the previous best/ powerful optical computer. Massively parallel arrays seem to be the future.
Go in AI research has benifitted by increasingly more powerful computing platforms and insights that can come from new uses of that power with FPGA’s that can be reprogrammed dynamically as they “learn”.
See also… TOP500.org lists of the most powerful conventional computers every 6 months…
SWitting on our old tech, without staying abreast of the bleeding edge is static, when the future designs are dynamic.
Advances in pattern sensing and AI are dependent on better computing platforms, algorithms, topologies and hardware as well as other factors, such as dynamic programming… See AMD’s purchase of Xilinx (FPGA’s) to be incorporated into newer generations of server chips for very large data centers and then being incorporated into engineering and powerful desktop systems for better data analysis in programs such as Tensorflow(neural networks).
Insights will also be applied to Go which has a very large complex of patterns, that need to be applied in the right location and timing with dynamic feedback… Isn’t that how you play Go at your best?
Why not stay upto date with processing capabilities that have emerged recently. I have been studying optical computers for over 30 years an percieved their value long ago. We are now starting to reach those levels.
Go is a large array, with complex analysis in only 19x19=361 locations. I have played 2-35x35 co-joined Go boards with 3 players per side in round robin sequence 30 years ago, to learn about how size, edges and center play are effected, as topology study. It was interesting how size effects playing styles. Be open minded and we can learn far more.
Leela/Alpha GO - Advanced AI Software to play AI Go games at a high level with large Neural Network size. AlphaGo as won against 4/5 games vs Lee Sedol-9p. Has cleaned house on 56? players on Korean Go server of all strong players the played AlphaGo-improved.
Combining Large Array Processing, Neural Networks(AI) on a large array Optical Computer is the current goal using current and bleeding edge tech, brought to bear on the game of Go.
Go game AI analysis continues to lead the field of AI.
I have chosen 2D arrays of x,y… ie.- a Go board taken to a much larger dimension… and then added other shapes and dimensions as needed for the Optical Computer…
…All this started with Go and the fact I am a programmer that has tried to think of other relationships (Go game shapes, board sizes and simplicity of the shapes of Go) and more complex levels of relationships… then AI, Neural Networking and more powerful computers/supercomputers and then simplifying the computing elements to a range of functional designs.
Please see if you can find the video of the recent Chinese (Dec 20,2020) in action…
It is a free space construction of 100x100 elements, (sound familiar… 19x19 only larger).
Large arrays can be made of 10 types of elements…
Corner, Side, Center, Laser emitters(arrays), Diode detctectors(arrays) or CCD’s linear or array and modulators, path switching and storage( for arithmetic logic, program and processing elements
The most complex element is the switching fabric by far…
All of which can be be arrays of elements w/ control logic.
Go has been the key and stimulus. Small, medium large arrays… or scaleable to any size with wafer fab and enough die real estate. Programable AI and Neural Network processing can be implemented by FPGA and ASICs(Application Specific Integrated Circuit) fixed logic but only FPGA dynamically reprogrammable logic.
I hope this helps you understand why Go and Optical Computing could lead to more breakthroughs like we have seen with strong AI Go programs and large storage patterns of games.
This is only one approach of a combination of elements and topologies and is not fixed in stone, but is a flexible design for the applications it is designed for. Analog Computing is also useful for weighting factors. The optimal Optical Computer would be both Analog and Digitalto optimize the design for the type of required computing functions needed.
These links are intended to open your minds to other possibilities, as in new Go move combinations and shapes(topologies)…
SAI copyright May 14,202, All Rights Reserved.