Skip to content

Pseudo-Lab/On-Device-AI_On-The-AIr

Folders and files

NameName
Last commit message
Last commit date

Latest commit

ย 

History

37 Commits
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 

Repository files navigation

On-Device AI: ON THE AIr

PseudoLab Discord Community Stars Badge Forks Badge Pull Requests Badge Issues Badge GitHub contributors

Welcome to On-Device AI: ON THE AIr repository! We aim to research On-Device AI with an emphasis on common model compression techniques, conducting paper reviews, and benchmarking real-world performance using NVIDIA Jetson devices. Join us in advancing On-Device AI through open collaboration and innovation! ๐Ÿš€

๐ŸŒŸ ํ”„๋กœ์ ํŠธ ๋ชฉํ‘œ (Project Vision)

"Propose the optimal model compression techniques for NVIDIA Jetson devices by leveraging the knowledge gained from research paper reviews on model compression methods."

  • Learn various pruning techniques during this season (10th).
  • Apply the learned model compression methods to existing models.
  • Test the actual performance on the NVIDIA Jetson platform.
  • Share the results for collaborative insights and community contribution.
  • Foster synergy between individual growth and collective intelligence.
  • Promote a knowledge-sharing culture based on the open-source spirit.

๐Ÿง‘ ์—ญ๋™์ ์ธ ํŒ€ ์†Œ๊ฐœ (Dynamic Team)

์—ญํ•  ์ด๋ฆ„ ๊ธฐ์ˆ  ์Šคํƒ ๋ฐฐ์ง€ ์ฃผ์š” ๊ด€์‹ฌ ๋ถ„์•ผ
Project Manager ์ •ํ˜„์šฐ Python PyTorch On-Device AI, CV, Robotics
Member ๊น€๋ฏผ์„ฑ Python -
Member ๊ตฌ์Šน์—ฐ Python -
Member ๋ฌธ๊ทœ์‹ Python -
Member ๋ฐ•์„ ์˜ Python -
Member ๋ฐ•์˜ˆ๋ฆฌ Python -
Member ์–‘๋ฌธ๊ธฐ Python -
Member ์ •์ง„์šฐ Python -
Member ์ตœ์˜ˆ์ œ Python -
Member ์ตœ์œ ์ง„ Python -
Member ์ตœํ•ด์ธ Python -

๐Ÿš€ ํ”„๋กœ์ ํŠธ ๋กœ๋“œ๋งต (Project Roadmap)

gantt
    title 2025 On-Device AI ํ”„๋กœ์ ํŠธ ์—ฌ์ •
    section ์ „์ฒด ์ปค๋ฆฌํ˜๋Ÿผ
    Pruning      :a1, 2025-03-03, 119d
    Quantization :a2, after a1, 120d

    section Pruning ์„ธ๋ถ€ ํ™œ๋™
    SPECIFIC OR UNIVERSAL SPEEDUP   :b1, 2025-03-03, 35d
    WHEN TO PRUNE                   :b2, after b1, 84d

    section ์‹ค์Šต ์„ธ๋ถ€ ํ™œ๋™ with Jetson
    Object Detection with Pruning   :c1, 2025-04-01, 63d
    LLM with Pruning                :c2, after c1, 30d
    CV with Pruning                 :c3, after c1, 30d
Loading

๐Ÿ’ป ์ฃผ์ฐจ๋ณ„ ํ™œ๋™ (Activity History)

Paper Review

๋‚ ์งœ ๋‚ด์šฉ ๋ฐœํ‘œ์ž ์ง„ํ–‰๋ฐฉ์‹ ์ฐธ๊ณ ์ž๋ฃŒ ๋น„๊ณ 
2025/03/05 OT ์ •ํ˜„์šฐ ์˜จ๋ผ์ธ -
2025/03/12 Unstructured Pruning ๊ตฌ์Šน์—ฐ ์˜จ๋ผ์ธ J. Frankle and M. Carbin, โ€œThe lottery ticket hypothesis: finding sparse, trainable neural networks,โ€ in ICLR, 2019.
2025/03/19 Structured Pruning ๊น€๋ฏผ์„ฑ ์˜คํ”„๋ผ์ธ X. Ma, G. Fang, and X. Wang, โ€œLLM-Pruner: On the structural pruning of large language models,โ€ in NeurIPS, vol. 36, 2023, pp.21 702โ€“21 720.
2025/03/26 Magical Week ํœด์ผ ๋ฏธ์ • - -
2025/04/03 Semi-structured Pruning ์ตœ์œ ์ง„ ์˜จ๋ผ์ธ F. Meng, H. Cheng, K. Li, H. Luo, X. Guo, G. Lu, and X. Sun, โ€œPruning filter in filter,โ€ in NeurIPSW, 2020.
2025/04/09 Pruning Before Training ๋ฌธ๊ทœ์‹ ์˜จ๋ผ์ธ S. Liu, T. Chen, X. Chen, L. Shen, D. C. Mocanu, Z. Wang, and M. Pechenizkiy, โ€œThe unreasonable effectiveness of random pruning: Return of the most naive baseline for sparse training,โ€ in ICLR, 2022.
2025/04/16 Pruning During Training: Sparsity Regularization based Methods ๋ฐ•์˜ˆ๋ฆฌ ์˜จ๋ผ์ธ W. Wen, C. Wu, Y. Wang, Y. Chen, and H. Li, โ€œLearning structured sparsity in deep neural networks,โ€ in NIPS, 2016.
2025/04/23 Pruning During Training: Dynamic Sparse Training based Methods ๊ตฌ์Šน์—ฐ ์˜คํ”„๋ผ์ธ U. Evci, T. Gale, J. Menick, P. S. Castro, and E. Elsen, โ€œRigging the lottery: Making all tickets winners,โ€ in ICML, 2020.
2025/04/30 Zero-shot Pruning ์ •ํ˜„์šฐ ์˜จ๋ผ์ธ ์„ ์ • ์ค‘ Magical Week
2025/05/07 Pruning During Training: Score-based Methods ์ตœํ•ด์ธ ์˜จ๋ผ์ธ Y. He, P. Liu, Z. Wang, Z. Hu, and Y. Yang, โ€œFilter pruning via geometric median for deep convolutional neural networks acceleration,โ€ in CVPR, 2019, pp. 4340โ€“4349.
2025/05/14 Pruning During Training: Differentiable Pruning based methods ์ •์ง„์šฐ ์˜จ๋ผ์ธ X. Ning, T. Zhao, W. Li, P. Lei, Y. Wang, and H. Yang, โ€œDSA: More efficient budgeted pruning via differentiable sparsity allocation,โ€ in ECCV, 2020, pp. 592โ€“607. Pseudo Con
2025/05/21 Pruning After Training: LTH and its Variants ์ •ํ˜„์šฐ ์˜จ๋ผ์ธ ์„ ์ • ์ค‘
2025/05/28 Pruning After Training: Other score-based Methods ๊น€๋ฏผ์„ฑ ์˜คํ”„๋ผ์ธ ์„ ์ • ์ค‘
2025/06/04 Pruning After Training: Sparsity Regularization based Methods ์ตœ์˜ˆ์ œ ์˜จ๋ผ์ธ ์„ ์ • ์ค‘
2025/06/11 Pruning After Training: Pruning in Early Training ์–‘๋ฌธ๊ธฐ ์˜จ๋ผ์ธ ์„ ์ • ์ค‘
2025/06/18 Pruning After Training: Post-Training Pruning ๋ฐ•์„ ์˜ ์˜จ๋ผ์ธ ์„ ์ • ์ค‘
2025/06/25 Run-time Pruning ์ •ํ˜„์šฐ ์˜คํ”„๋ผ์ธ ์„ ์ • ์ค‘

Hands-On Pruning with Jetson

๋‚ ์งœ ๋‚ด์šฉ ์ง„ํ–‰๋ฐฉ์‹ ๋น„๊ณ 
2025/04/01 OT ๋ฐ ๊ณ„ํš ์ˆ˜๋ฆฝ ์˜จ๋ผ์ธ
2025/04/15 Object Detection Model ์„ ์ • ์˜จ๋ผ์ธ
2025/04/29 [PDT] Sparsity Regularization based Method ๊ตฌํ˜„ ๋ฐ ํ…Œ์ŠคํŠธ ์˜จ๋ผ์ธ Magical Week
2025/05/06 ASP ๊ธฐ๋ฐ˜ ๋ชจ๋ธ ํ•™์Šต ์˜จ๋ผ์ธ
2025/05/13 TensorRT ๋ณ€ํ™˜ ๋ฐ HW ๋‚ด ์„ฑ๋Šฅ ๋น„๊ต ์˜คํ”„๋ผ์ธ PseudoCon
2025/05/20 [PDT] Sparse Training based Methods ๊ตฌํ˜„ ๋ฐ ํ…Œ์ŠคํŠธ ์˜จ๋ผ์ธ
2025/05/27 [PDT] Score-based Methods ๊ตฌํ˜„ ๋ฐ ํ…Œ์ŠคํŠธ ์˜จ๋ผ์ธ
2025/06/03 [PDT] Differentiable Pruning based methods ๊ตฌํ˜„ ๋ฐ ํ…Œ์ŠคํŠธ ์˜จ๋ผ์ธ
2025/06/10 [PDT] ๊ตฌํ˜„๋œ ๋ชจ๋ธ๋“ค TensorRT ๋ณ€ํ™˜ ๋ฐ HW ์„ฑ๋Šฅ ๋น„๊ต ์˜คํ”„๋ผ์ธ
2025/06/17 [PAT] LTH and its Variants ๊ตฌํ˜„ ๋ฐ ํ…Œ์ŠคํŠธ ์˜จ๋ผ์ธ
2025/06/24 [PAT] Pruning in Early Training ๊ตฌํ˜„ ๋ฐ ํ…Œ์ŠคํŠธ ์˜จ๋ผ์ธ
2025/07/01 [PAT] Post-Training Pruning ๊ตฌํ˜„ ๋ฐ ํ…Œ์ŠคํŠธ ์˜จ๋ผ์ธ
2025/07/08 Run-time Pruning ๊ตฌํ˜„ ๋ฐ ํ…Œ์ŠคํŠธ ์˜จ๋ผ์ธ

์ง„ํ–‰ ๋ฐฉ์‹

Paper Review

๋งค์ฃผ ์Šคํ„ฐ๋”” ์ง„ํ–‰ ๋ฐฉ์‹์€ ๋‹ค์Œ๊ณผ ๊ฐ™์Šต๋‹ˆ๋‹ค.

  1. ๊ทผํ™ฉ ์ด์•ผ๊ธฐ (20 ~ 30๋ถ„ ์˜ˆ์ƒ)
  2. ๋ฐœํ‘œ์ž๋ฅผ ์ œ์™ธํ•œ ์ฐธ์—ฌ์ž๋“ค์ด ์ค€๋น„ํ•œ On-Device AI ๊ด€๋ จ๋œ ์ด์Šˆ๋“ค์„ ๊ณต์œ ํ•œ๋‹ค. (20 ~ 40๋ถ„ ์˜ˆ์ƒ)
  3. ๋ฐœํ‘œ์ž๋Š” ์ค€๋น„ํ•œ ๋…ผ๋ฌธ ๋ฆฌ๋ทฐ๋ฅผ ๋ฐœํ‘œํ•œ๋‹ค. (30๋ถ„ ~ 1์‹œ๊ฐ„ ์˜ˆ์ƒ)

์ด์— ๋”ฐ๋ผ ๋‹ค์Œ ๋‚ด์šฉ๋“ค์„ ์ค€๋น„ํ•˜์‹œ๋ฉด ๋ฉ๋‹ˆ๋‹ค
๊ณตํ†ต์‚ฌํ•ญ

  • ํ•ด๋‹น ์ฃผ์ฐจ ๋…ผ๋ฌธ์„ ์ฝ๋Š”๋‹ค.

๋ฐœํ‘œ์ž

  • ํ•ด๋‹น ์ฃผ์ฐจ ๋…ผ๋ฌธ์— ๋Œ€ํ•œ ๋ฐœํ‘œ ์ค€๋น„๋ฅผ ํ•œ๋‹ค.

์ฐธ์—ฌ์ž

  • On-Device AI์™€ ๊ด€๋ จ๋œ ๊ธฐ์ˆ ๋“ค(TensorRT, LiteRT, ONNX ๋“ฑ)์˜ ํŠธ๋ Œ๋“œ๋‚˜ ์ด์Šˆ๋ฅผ ์ค€๋น„ํ•œ๋‹ค.

๐Ÿ’ก ํ•™์Šต ์ž์› (Learning Resources)

์„ธ๋ถ€ ๋…ผ๋ฌธ๋“ค์€ ์ฃผ์ฐจ๋ณ„ ํ™œ๋™ ๋‚ด ์ฐธ๊ณ ์ž๋ฃŒ ์ฐธ๊ณ 

์ฐธ๊ณ  ๋ฌธํ—Œ

๐ŸŒฑ ์ฐธ์—ฌ ์•ˆ๋‚ด (How to Engage)

์ง„ํ–‰ ์ •๋ณด

  • ์‹œ๊ฐ„: ๋งค์ฃผ ์ˆ˜์š”์ผ ์˜คํ›„ 8์‹œ
  • ์žฅ์†Œ: ์˜จ๋ผ์ธ / ์˜คํ”„๋ผ์ธ(๊ฐ•๋‚จ์—ญ)

์ฐธ์—ฌ ์กฐ๊ฑด

  • On-Device AI(๊ฒฝ๋Ÿ‰ํ™”, ์ตœ์ ํ™” ๋“ฑ)์— ๊ด€์‹ฌ ์žˆ์œผ์‹  ๋ถ„
  • 4๊ฐœ์›” ๋™์•ˆ ๊พธ์ค€ํžˆ ์ฐธ์—ฌํ•˜์‹ค ์ˆ˜ ์žˆ๋Š” ๋ถ„
  • ๋”ฅ๋Ÿฌ๋‹ ๊ธฐ์ดˆ ์ง€์‹ ๋ณด์œ ํ•˜์‹  ๋ถ„
  • ๋…ผ๋ฌธ์„ ์ฝ๊ณ  ๋ฆฌ๋ทฐํ•˜์‹ค ์ˆ˜ ์žˆ๋Š” ๋ถ„

ํŒ€์›์œผ๋กœ ์ฐธ์—ฌํ•˜์‹œ๋ ค๋ฉด ๋Ÿฌ๋„ˆ ๋ชจ์ง‘ ๊ธฐ๊ฐ„์— ์‹ ์ฒญํ•ด์ฃผ์„ธ์š”.

  • ๋งํฌ (์ค€๋น„์ค‘)

๋ˆ„๊ตฌ๋‚˜ ์ฒญ๊ฐ•์„ ํ†ตํ•ด ๋ชจ์ž„์„ ์ฐธ์—ฌํ•˜์‹ค ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

  1. ํŠน๋ณ„ํ•œ ์‹ ์ฒญ ์—†์ด ์ •๊ธฐ ๋ชจ์ž„ ์‹œ๊ฐ„์— ๋งž์ถ”์–ด ๋””์Šค์ฝ”๋“œ #Room-GH ์ฑ„๋„๋กœ ์ž…์žฅ
  2. Magical Week ์ค‘ ํ–‰์‚ฌ์— ์ฐธ๊ฐ€
  3. Pseudo Lab ํ–‰์‚ฌ์—์„œ ๋งŒ๋‚˜๊ธฐ

About Pseudo Lab ๐Ÿ‘‹๐Ÿผ

Pseudo-Lab is a non-profit organization focused on advancing machine learning and AI technologies. Our core values of Sharing, Motivation, and Collaborative Joy drive us to create impactful open-source projects. With over 5k+ researchers, we are committed to advancing machine learning and AI technologies.

Contributors ๐Ÿ˜ƒ



License ๐Ÿ—ž

This project is licensed under the MIT License.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 4

  •  
  •  
  •  
  •  

Languages