BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//cfp.riscv-europe.org//eu-summit-2026//speaker//D3YWSP
BEGIN:VTIMEZONE
TZID:CET
BEGIN:STANDARD
DTSTART:20001029T040000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:CET
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000326T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:CEST
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-eu-summit-2026-SRK9TJ@cfp.riscv-europe.org
DTSTART;TZID=CET:20260610T153000
DTEND;TZID=CET:20260610T160000
DESCRIPTION:This demo showcases complete end-to-end Transformer training lo
 cally on the GAP9 RISC-V MCU. On-device training is crucial for applicatio
 ns that operate in dynamically changing environments. One example is biosi
 gnal DNNs in wearable devices\, where cross-subject transfer and long-term
  temporal drift degrade performance. RISC-V MCUs are already widely used f
 or edge DNN deployment. However\, most existing work focuses either on inf
 erence only\, or on fine-tuning a small portion of the network.\n\nWe exte
 nded the Deeploy compiler to generate training code. Deeploy generates bar
 e-metal C code from an ONNX graph and is tailored for efficient inference.
  To support training\, we added critical kernels such as optimizers and in
 -place gradient accumulators. We also extended the ONNX runtime training A
 PI to generate graphs optimized for edge deployment. This extension is rel
 eased at https://github.com/pulp-platform/ONNX4Deeploy. To reduce the memo
 ry footprint of batching required for stable training\, we implement gradi
 ent accumulation. The demo video showcases the full workflow\, from traini
 ng graph optimization to code generation and on-board execution. The video
  is available at https://drive.google.com/file/d/16BMiHn0jyMvScFJD7AGTwHpA
 4Rc0aMnC/view?usp=drive_link and will be uploaded to the Pulp Platform You
 Tube channel.
DTSTAMP:20260522T162432Z
LOCATION:Devzone
SUMMARY:End-to-End On-Device Transformer Training on Ultra-Low Power RISC-V
  MCU - Victor Jung\, RunW
URL:https://cfp.riscv-europe.org/eu-summit-2026/talk/SRK9TJ/
END:VEVENT
END:VCALENDAR
