Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
80 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
7 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

From Cloud to Edge: Rethinking Generative AI for Low-Resource Design Challenges (2402.12702v2)

Published 20 Feb 2024 in cs.AI and cs.CY

Abstract: Generative AI has shown tremendous prospects in all aspects of technology, including design. However, due to its heavy demand on resources, it is usually trained on large computing infrastructure and often made available as a cloud-based service. In this position paper, we consider the potential, challenges, and promising approaches for generative AI for design on the edge, i.e., in resource-constrained settings where memory, compute, energy (battery) and network connectivity may be limited. Adapting generative AI for such settings involves overcoming significant hurdles, primarily in how to streamline complex models to function efficiently in low-resource environments. This necessitates innovative approaches in model compression, efficient algorithmic design, and perhaps even leveraging edge computing. The objective is to harness the power of generative AI in creating bespoke solutions for design problems, such as medical interventions, farm equipment maintenance, and educational material design, tailored to the unique constraints and needs of remote areas. These efforts could democratize access to advanced technology and foster sustainable development, ensuring universal accessibility and environmental consideration of AI-driven design benefits.

PDF HTML Abstract

Rethinking Generative AI Deployment for Low-Resource Settings

Introduction to Edge AI in Design

The progressive advancements in Generative AI have ushered in a new era in the field of design, with applications spanning across medical, agricultural, and educational sectors. The core premise of the discussed paper focuses on the pivotal shift from traditional cloud-based Generative AI applications to implementations that cater to edge computing—particularly emphasizing scenarios characterized by limited computational resources. This shift is crucial for democratizing AI, ensuring its benefits reach remote and resource-constrained areas, thereby promoting sustainable development and universal accessibility.

Challenges and Potential Solutions

The paper acknowledges the significant hurdles in adapting complex Generative AI models for efficient operation within low-resource settings. These challenges stem from the inherent need for substantial memory, computational power, and continuous internet connectivity—requirements that are far from being met in remote areas. To address these challenges, the paper advocates for innovative approaches in model compression, algorithmic efficiency, and the potential use of edge computing architectures. Key strategies include:

  • Model Compression: Techniques such as model pruning, quantization, and knowledge distillation are highlighted as potential methods to reduce the complexity and size of generative models, facilitating deployment on devices with limited capacities.
  • Edge Computing: The paper suggests leveraging edge computing to process data closer to its source, thereby reducing reliance on high-speed internet connections and large data centers, which are scarce in remote locations.

Review of Current Models and Their Limitations

The paper offers a comprehensive review of current Generative AI models, including LLMs, Diffusion Models, and Vision LLMs (VLMs). It underscores the limitations these models face when deployed in resource-constrained environments, mainly due to their extensive computational and memory requirements. Despite their proven capabilities in generating human-like text and high-quality images, the models' dependability on substantial computational resources limits their applicability in remote areas.

Implications for the Developing World

A significant portion of the paper is devoted to the implications of deploying offline, lightweight AI models in developing countries. By focusing on specific case studies in the fields of medical intervention, farm equipment maintenance, and the design of educational materials, the paper elucidates how tailored AI solutions can address unique local challenges. The deployment of offline models trained on locally relevant data promises more tailored, effective, and sustainable design solutions that align with the ecological sensitivities and material constraints prevalent in these regions.

Future Directions in AI for Design

The paper speculates on the future advancements in AI-driven design, emphasizing the need for continued research in model optimization, hardware innovations, and the development of new tools and methodologies for leveraging AI at the edge. It stresses the importance of making Generative AI models more accessible and highlights the potential impact of TinyML and Edge AI as facilitators of this transformative shift. The envisioned future is one where AI-driven design tools are universally accessible, ensuring that the benefits of technological advancements are equitably distributed.

Conclusion

In encapsulating the discussion, the paper reaffirms the necessity of rethinking Generative AI deployment for resource-constrained settings. It calls for a coordinated effort among researchers, developers, and policymakers to innovate and develop solutions that bridge the accessibility gap in AI technology. By shifting the focus from cloud-based to edge-based applications, the paper envisions a future where AI-driven design benefits those in the most remote corners of the globe, fostering equitable progress and sustainable development.

File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown Bookmark Chat Bubble Oval Streamline Icon: https://streamlinehq.com Chat (Pro)
Definition Search Book Streamline Icon: https://streamlinehq.com
References (42)
  1. Flamingo: a visual language model for few-shot learning. Advances in Neural Information Processing Systems, 35: 23716–23736.
  2. BriteLab, P. 2023. Greenhouse Grid Design Task. Accessed: 11-28-2023.
  3. Artificial Intelligence-Based Fault Diagnosis and Prediction for Smart Farm Information and Communication Technology Equipment. Agriculture, 13(11): 2124.
  4. LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale. arXiv:2208.07339.
  5. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. arXiv:1810.04805.
  6. Tinyml meets iot: A comprehensive survey. Internet of Things, 16: 100461.
  7. A Survey of Quantization Methods for Efficient Neural Network Inference. arXiv:2103.13630.
  8. ImageBind: One Embedding Space To Bind Them All. In CVPR.
  9. Knowledge Distillation of Large Language Models. arXiv:2306.08543.
  10. Distilling Step-by-Step! Outperforming Larger Language Models with Less Training Data and Smaller Model Sizes. arXiv:2305.02301.
  11. Jain, M. 2023. Center for Creative Learning - Toys and Activities. In https://www.ccl.iitgn.ac.in/toys.
  12. Developing world users as lead users: a case study in engineering reverse innovation. Journal of Mechanical Design, 137(7): 071406.
  13. Techology trend of edge AI. In 2018 International Symposium on VLSI Design, Automation and Test (VLSI-DAT), 1–2.
  14. Pushing Large Language Models to the 6G Edge: Vision, Challenges, and Opportunities. arXiv:2309.16739.
  15. LLM-QAT: Data-Free Quantization Aware Training for Large Language Models. arXiv:2305.17888.
  16. LLM-Pruner: On the Structural Pruning of Large Language Models. arXiv:2305.11627.
  17. Simulating the Adoption and Social Impact of Improved Cookstoves in Uganda Using Agent-Based Modeling and Neural Networks. Journal of Mechanical Design, 145(12).
  18. Edge AI: Systems Design and ML for IoT Data Analytics. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, KDD ’20, 3565–3566. New York, NY, USA: Association for Computing Machinery. ISBN 9781450379984.
  19. Why the developing world needs mechanical design. Journal of Mechanical Design, 138(7): 070301.
  20. Nine principles for design for the developing world as derived from the engineering literature. Journal of Mechanical Design, 136(12): 121403.
  21. Machine Learning at the Network Edge: A Survey. ACM Computing Surveys, 54(8): 1–37.
  22. OpenAI. 2023. GPT-4 Technical Report. arXiv:2303.08774.
  23. Training language models to follow instructions with human feedback. arXiv:2203.02155.
  24. From Concept to Manufacturing: Evaluating Vision-Language Models for Engineering Design. arXiv:2311.12668.
  25. SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis. arXiv:2307.01952.
  26. Is TinyML Sustainable? Assessing the Environmental Impacts of Machine Learning on Microcontrollers. arXiv preprint arXiv:2301.11899.
  27. Zero-Shot Text-to-Image Generation. arXiv:2102.12092.
  28. Ray, P. P. 2022. A review on TinyML: State-of-the-art and prospects. Journal of King Saud University-Computer and Information Sciences, 34(4): 1595–1623.
  29. Deep generative models in engineering design: A review. Journal of Mechanical Design, 144(7): 071704.
  30. Beyond statistical similarity: Rethinking metrics for deep generative models in engineering design. arXiv preprint arXiv:2302.02913.
  31. High-Resolution Image Synthesis with Latent Diffusion Models. arXiv:2112.10752.
  32. DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter. arXiv:1910.01108.
  33. DesIGN: Design Inspiration from Generative Networks. In Proceedings of the European Conference on Computer Vision (ECCV) Workshops.
  34. Edge AI: a survey. Internet of Things and Cyber-Physical Systems.
  35. Combining direct and indirect user data for calculating social impact indicators of products in developing countries. Journal of Mechanical Design, 142(12): 121401.
  36. A Simple and Effective Pruning Approach for Large Language Models. arXiv:2306.11695.
  37. Any-to-Any Generation via Composable Diffusion. arXiv preprint arXiv:2305.11846.
  38. LLaMA: Open and Efficient Foundation Language Models. Cite arxiv:2302.13971.
  39. Assessing the impact of generative AI on medicinal chemistry. Nature biotechnology, 38(2): 143–145.
  40. Chain-of-Thought Prompting Elicits Reasoning in Large Language Models. arXiv:2201.11903.
  41. Design for the developing world: Common pitfalls and how to avoid them. Journal of Mechanical Design, 138(3): 031101.
  42. Visual ChatGPT: Talking, Drawing and Editing with Visual Foundation Models. arXiv:2303.04671.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (5)
  1. Sai Krishna Revanth Vuruma (4 papers)
  2. Ashley Margetts (1 paper)
  3. Jianhai Su (4 papers)
  4. Faez Ahmed (66 papers)
  5. Biplav Srivastava (57 papers)
Citations (3)
View on Semantic Scholar
X Twitter Logo Streamline Icon: https://streamlinehq.com

Tweets