![[PukiWiki]](image/pukiwiki.png "[PukiWiki]")
#author("2024-11-05T20:09:32+09:00","default:asl_member","asl_member") #author("2024-11-05T20:13:44+09:00","default:asl_member","asl_member") -[[Members-Internal]] ---- CENTER:COLOR(#81111F){SIZE(40){''Dynamic Quantization & Pruning in Spiking Neuron Networks''}} ---- *Background [#d3104867] As Moore's Law approaches its physical limitations, the demand for alternative computing architectures grows. Neuromorphic computing, particularly Spiking Neuron Networks (SNNs), stands out due to its ability to mimic the brain's energy-efficient, event-driven processing. SNNs excel in temporal data processing and are particularly well-suited for applications requiring low power and real-time computation. This research focuses on dynamic quantization and pruning methods, which aim to reduce the computational complexity and energy consumption of SNNs without significantly sacrificing accuracy. These techniques offer the potential to unlock even greater efficiency, making SNNs more viable for widespread adoption in edge computing and embedded systems. *Motivation [#w893ae2e] Neuromorphic computing, especially SNNs, has shown great promise in areas requiring high power efficiency and rapid data processing. Prior research in static quantization and pruning has demonstrated significant improvements in computational performance and energy savings, but these approaches do not adapt dynamically to varying workloads or real-time conditions. There is a clear need for methods tailored specifically for SNNs that can adjust on the fly. By developing and integrating dynamic quantization and pruning techniques, this research aims to bridge that gap and push the boundaries of power-efficient computing in neuromorphic systems. *Goal [#v20a6f9f] -Develop and evaluate novel dynamic quantization and pruning methods for SNNs. -Optimize both energy consumption and computational efficiency while maintaining or improving accuracy. -Build a comprehensive evaluation framework to rigorously test these methods across various benchmarks. -Contribute to the advancement of neuromorphic computing by creating more power-efficient neural network models. *Research Plan [#bc523092] ***Step 1: [#u6cd39e3] 1- Read and gain knowledge about neuromorphic systems. 2- Guidance tutorials ***Step 2: [#n00c8de5] *Research Schedule [#r3e361fe] |Date|Task| |April - July 2024 | Field exploration | |August - September 2024 |Learning about Hardware| |September 30 - October 04 2024 | Tutorials | |October 04 - October 11 2024 | Testing SNN power estimation — &ref(Report1.pdf); | |October 14 - October 28 2024 | Preparing figures and content of the 'White-paper' &ref(Report2.pdf); | |October 14 - October 28 2024 | Preparing figures and content of the 'White-paper' — &ref(Report2.pdf); | |October 28 - ~ | Understanding the 'ODIN' processor &ref(https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=8528875); | *Availability[#g0c8f9db] | Monday | Tuesday | Wednesday | Thursday | Friday | | In Lab | In Lab | X | X | In Lab |
