A new research paper explores the impact of temperature on analog deep neural network (DNN) inference, particularly for resource-constrained devices like mobile phones. The study found that temperature significantly degrades inference accuracy, primarily due to systematic non-idealities rather than just stochastic noise. Mitigation strategies such as noise-aware training, hardware-in-the-loop training, and temperature-aware calibration were evaluated, with the latter two proving most effective at maintaining accuracy under varying thermal conditions. AI
IMPACT This research could lead to more robust and energy-efficient AI deployments on edge devices by addressing environmental performance degradation.
RANK_REASON Research paper published on arXiv detailing experimental study on analog DNN inference. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- deep neural network
- Embedded Devices
- hardware-in-the-loop training
- mobile phone
- Noise-aware training
- temperature-aware calibration
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
