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:20260402T024534Z
LOCATION:3000\, Level 3
DTSTART;TZID=America/Los_Angeles:20250624T103000
DTEND;TZID=America/Los_Angeles:20250624T104500
UID:dac_DAC 2025_sess116_RESEARCH2288@linklings.com
SUMMARY:NSFlow: An End-to-End FPGA Framework with Scalable Dataflow Archit
 ecture for Neuro-Symbolic AI
DESCRIPTION:Hanchen Yang and Zishen Wan (Georgia Institute of Technology);
  Ritik Raj (Student); and Joongun Park, Ziwei Li, Ananda Samajdar, Arijit 
 Raychowdhury, and Tushar Krishna (Georgia Institute of Technology)\n\nAbst
 ract—Neuro-Symbolic AI (NSAI) is an emerging paradigm that integrates neur
 al networks with symbolic reasoning to enhance the trans-parency, reasonin
 g capabilities, and data efficiency of AI systems. Recent NSAI systems hav
 e gained traction due to their exceptional performance in reasoning tasks 
 and human-AI collaborative scenarios. Despite these algorithmic advancemen
 ts, executing NSAI tasks on existing hardware (e.g., CPUs, GPUs, TPUs) rem
 ains challenging, due to their heterogeneous computing kernels, high memor
 y intensity, and unique memory access patterns. Moreover, current NSAI alg
 orithms exhibit significant variation in operation types and scales, makin
 g them incompatible with existing ML accelerators. These challenges highli
 ght the need for a versatile and flexible acceleration framework tailored 
 to NSAI workloads.\nIn this paper, we propose NSFlow, an FPGA-based accele
 ration framework designed to achieve high efficiency, scalability, and ver
 satility across NSAI systems. NSFlow features a design architecture genera
 -\ntor that identifies workload data dependencies and creates optimized da
 taflow architectures, as well as a reconfigurable array with flexible comp
 ute units, re-organizable memory, and mixed-precision capabilities. Evalua
 ting across NSAI workloads, NSFlow achieves 31× speedup over Jetson TX2, m
 ore than 2× over GPU, 8× speedup over TPU-like systolic array, and more th
 an 3× over Xilinx DPU. NSFlow also demonstrates enhanced scalability, with
  only 4× runtime increase when symbolic workloads scale by 150×. To the be
 st of our knowledge, NSFlow is the first framework to enable real-time gen
 eralizable NSAI algorithms acceleration, demonstrating a promising solutio
 n for next-generation cognitive systems.\n\nTopics: AI\n\nTracks: AI4: AI/
 ML System and Platform Design\n\nSession Chairs: Kai-Yuan (Kevin) Chao (Si
 emens) and Amin Firoozshahian (Rain AI)\n\n
END:VEVENT
END:VCALENDAR
