InstantFT fine-tunes a LeNet-5-like CNN on a Xilinx Kria KV260 FPGA in 0.36 seconds, with accuracy close to LoRA-All, by caching frozen features and training low-rank adapters that connect every layer to the output.
Hero III, and Pramod K
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
InstantFT: An FPGA-Based Runtime Subsecond Fine-tuning of CNN Models
InstantFT fine-tunes a LeNet-5-like CNN on a Xilinx Kria KV260 FPGA in 0.36 seconds, with accuracy close to LoRA-All, by caching frozen features and training low-rank adapters that connect every layer to the output.