BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Los_Angeles
X-LIC-LOCATION:America/Los_Angeles
BEGIN:DAYLIGHT
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
TZNAME:PDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
TZNAME:PST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260730T152640Z
LOCATION:Exhibit Hall
DTSTART;TZID=America/Los_Angeles:20260728T173800
DTEND;TZID=America/Los_Angeles:20260728T173900
UID:dac_DAC 2026_sess306_LBR054@linklings.com
SUMMARY:Late Breaking Results: ASTFusion -- Two-Stage Structural Enhanceme
 nt Learning  for Robust HLS Code Generation
DESCRIPTION:Ke Zhiwei (University of Science&Technology of China); Wenqi L
 ou (USTC); Liu Yiming, Fengrui Zuo, and Chao Wang (University of Science a
 nd Technology of China); and Xuehai Zhou (USTC)\n\nRecent studies show tha
 t abstract syntax trees (ASTs) can improve the syntactic validity of large
  language model (LLM)-generated high-level synthesis (HLS) code. However, 
 structure-enhanced methods often degrade at inference due to reliance on e
 xplicit structural signals during training. We show that robustness depend
 s on how structure is introduced, rather than simply whether it is used. W
 e propose ASTFusion, a learn--then--utilize framework that decouples struc
 tural acquisition from structural conditioning. ASTFusion internalizes hie
 rarchical and dependency-aware structure through reconstruction from incom
 plete implementations, and exploits partial structural cues as optional gu
 idance during generation. On HLSEval, ASTFusion improves Func@1 from 40.00
 % to 72.34% over standard HLS fine-tuning and from 57.69% to 72.34% over p
 rior AST-enhanced fine-tuning, while reaching 98.71% Synth@10. These resul
 ts show that progressive structural enhancement is effective for robust HL
 S code generation under realistic inference constraints.\n\nTrack: Student
 \n\n
END:VEVENT
END:VCALENDAR