Pith. sign in

REVIEW 3 major objections 4 minor 34 references

TRINE: A Token-Aware, Runtime-Adaptive FPGA Inference Engine for Multimodal AI

T0 review · 3 major / 4 minor · reviewed 2026-07-13 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read One FPGA bitstream can run full multimodal AI (vision, language, graphs) at low power by switching dataflows and pruning tokens on the fly.

desk verdict Solid single-bitstream multimodal FPGA engine with real measured gains; the 22.57× GPU number is batch-1 and should not be the headline, but the architecture and FPGA results still stand. read the letter →

arxiv 2603.22867 v1 pith:HRSZPOD5 submitted 2026-03-24 cs.AR cs.AIcs.LG

Hyunwoo Oh , Hanning Chen , Sanggeon Yun , Yang Ni , Suyeon Jang , Behnam Khaleghi , Fei Wen , Mohsen Imani This is my paper · ORCID
classification cs.ARcs.AIcs.LG
keywords FPGAacceleratormultimodalinferencetokenpruningmode-switchablePEarraysparsematrixkernelsViTCNNGNNNLPruntimeadaptivedataflowsinglebitstream
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Multimodal models that mix vision transformers, CNNs, graph networks, and language transformers pull embedded systems in different directions: dense matrix multiply one moment, sparse reductions the next, with real-time deadlines that leave little room for reconfiguration. This paper argues that those layers can be rewritten as three matrix kernels and executed end-to-end on a single FPGA bitstream. The hardware reuses one processing-element array that switches at runtime among systolic, SIMD, and sparse-reduction modes, prunes low-importance tokens as scores are produced, and overlaps independent kernels across multiple units. On commercial FPGAs the design reports large latency cuts versus a high-end GPU and an embedded GPU while drawing only about 20 W, with quantization keeping accuracy losses small. A sympathetic reader cares because the result is a practical path to hard-real-time multimodal inference without swapping bitstreams or shipping separate accelerators for each modality.

What carries the argument

The mode-switchable engine (MSE): a shared PE array whose interconnect and per-PE ops reconfigure at runtime among systolic, 1xCS SIMD, and RADT dataflows, fed by a width-matched two-stage top-k unit and scheduled by DALO across an RPU grid.

What would settle it

Re-run the same TinyCLIP, MDETR, and MissionGNN end-to-end graphs at batch 1 with identical int8 models and pruning rates on the cited FPGA baselines and on the RTX 4090 / Orin Nano without transfer-overlap assumptions; if TRINE’s latency and energy advantages disappear or reverse under those matched conditions, the central claim fails.

Watch

Extended reading notes

Core claim

TRINE shows that ViT, CNN, GNN, and transformer-NLP layers can be unified as DDMM, SDDMM, and SpMM and mapped onto one mode-switchable PE array that toggles among weight/output-stationary systolic, 1xCS SIMD, and a routable adder tree, combined with in-stream top-k token pruning and dependency-aware layer offloading, so end-to-end multimodal inference runs on a single bitstream without reconfiguration while delivering large measured latency and energy gains.

Load-bearing premise

That the DSP-normalized and FLOPs-scaled comparisons to single-modal FPGA designs, plus batch-1 GPU baselines with the chosen pruning rate, fairly represent real deployment conditions rather than favoring TRINE.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. TRINE is a single-bitstream FPGA accelerator and compiler for end-to-end multimodal inference (ViT, CNN, GNN, transformer NLP) without reconfiguration. Layers are unified as DDMM/SDDMM/SpMM and mapped to a mode-switchable engine (MSE) that runtime-toggles among weight/output-stationary systolic, 1×CS SIMD, and a routable adder tree (RADT) on a shared PE array. A width-matched two-stage top-k unit supports in-stream token pruning, and dependency-aware layer offloading (DALO) overlaps independent kernels across an RPU grid. On Alveo U50 and ZCU104 (int8), the authors report up to 22.57× lower latency than an RTX 4090 and 6.86× than a Jetson Orin Nano at ~20–21 W, with pruning alone up to 7.8× on ViT-heavy cases, DALO up to 79% throughput gain, and accuracy drops <2.5%.

Significance. If the claims hold under fair baselines, TRINE is a meaningful systems contribution: a practical single-bitstream substrate that covers four modalities with runtime dataflow switching, hardware top-k pruning, and inter-kernel overlap. The architecture (shared PE array + RADT + in-stream sorter + DALO) is concrete and well-motivated relative to prior single-modal or reconfiguration-heavy FPGA designs. Resource tables, end-to-end latencies with/without DALO, pruning ablations, and int8 accuracy numbers on two platforms give a solid empirical core. The work is relevant to embedded multimodal inference where reconfiguration cost and heterogeneous kernels are real constraints.

major comments (3)
  1. [§5.4.2, Table 5] Table 5 and §5.4.2: the headline 22.57× (TinyCLIP-A: U50 1.6 ms vs RTX 4090 37.1 ms) is measured at batch size 1 with transfers overlapped. For a small pruned ViT+text model this regime is expected to leave a high-end GPU occupancy-starved; the paper reports no GPU occupancy, kernel-fusion status, or batch-size sweep. The same table already shows the gap collapsing or reversing on larger configs (MDETR ~1.2×, MissionGNN-J 0.88×), consistent with utilization effects. Without those controls, the SOTA latency claim does not generalize to the deployment regimes the abstract advertises and should be restated with batch-size sensitivity or a utilization-aware GPU baseline.
  2. [§5.4.1, Table 4] Table 4 and §5.4.1: several competing FPGA numbers are FLOPs-scaled estimates or DSP-normalized latencies (footnotes). Normalization by DSP fraction credits designs that use fewer DSPs, but does not establish wall-clock superiority on comparable silicon, and FLOPs scaling for single-modal baselines (e.g., HeatViT on TinyCLIP) is not validated against measured multimodal end-to-end runs. The claim that TRINE is the only single-bitstream end-to-end CNN+ViT+GNN+NLP engine is plausible, but the quantitative SOTA ranking over prior FPGAs needs either measured apples-to-apples runs or a clearer separation of architectural novelty from normalized estimates.
  3. [§5.1–5.2, Table 5] §5.1–5.2 and Fig. 5–6: MissionGNN-J (ViT-H/14) is 414 ms without pruning and still 56.9 ms with p=0.3 on U50—slower than the RTX 4090 in Table 5—while the abstract leads with the best-case 22.57×. The evaluation should more clearly bound when TRINE wins (small/medium pruned ViT+text, graph-heavy SpMM) versus when FPGA capacity, not dataflow, is the limiter, so the central efficiency claim is not overstated by the best single multiplier.
minor comments (4)
  1. [§2, Table 1] Table 1 and related-work discussion: several prior works are marked as lacking run-time pruning or full modality coverage; a short note on whether any of them could be composed (e.g., VisionAGILE + a pruning unit) would strengthen the uniqueness claim without changing the architecture contribution.
  2. [§3.2, Fig. 1] Fig. 1 and §3.2: mode-selection policy (WS vs OS; 1×CS vs RADT) is described qualitatively by shape and sparsity; a short decision table or pseudocode would make the compiler/runtime policy reproducible.
  3. [Table 2, §5.4] Power is reported as board-level ~21 W (U50) and 8 W (ZCU104) with no measurement methodology (on-chip sensors vs wall power, idle vs active). A one-sentence method note would help energy comparisons in Tables 4–5.
  4. [Abstract, Fig. 5] Minor presentation: abstract and intro use both "1xCS" and "1×CS"; keep notation consistent. A few figure panels (Fig. 5) are dense; axis labels for with/without DALO could be clearer.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation; claims are measured end-to-end FPGA results against external GPU and FPGA baselines.

full rationale

TRINE is a systems/architecture paper. Its load-bearing claims (single-bitstream multimodal execution, up to 22.57× vs RTX 4090, 6.86× vs Orin Nano, pruning up to 7.8×, DALO up to 79 %, int8 accuracy drop <2.5 %) are obtained by RTL synthesis, place-and-route, and wall-clock measurement on Alveo U50/ZCU104 (Tables 2–5, Figs. 5–7), not by algebraic derivation from fitted parameters. Layer unification as DDMM/SDDMM/SpMM and the MSE mode-selection policy are explicit design choices whose benefit is shown empirically; they do not reduce a “prediction” to an input by construction. Self-citations (MissionGNN, TaskCLIP) supply evaluation workloads whose authors partially overlap, but those citations are not used as uniqueness theorems or load-bearing premises for the accelerator results; the results stand on third-party hardware comparisons. No equation equates a claimed output to a fitted quantity, no ansatz is smuggled via self-citation, and no known empirical pattern is merely renamed. Per the hard rules this is an honest non-finding: score 0, empty steps.

Assumptions & free parameters 4 free parameters · 4 assumptions · 4 invented entities

Hardware papers rest on design choices and evaluation conventions rather than mathematical axioms. Free parameters are the concrete micro-architectural knobs and pruning rates chosen for the reported builds; axioms are standard FPGA and sparse-linear-algebra assumptions; invented entities are the named architectural blocks introduced by the work.

free parameters (4)
  • PE array dimensions (RS × CS) = 32×32
    Fixed at 32×32 for both U50 and ZCU104 builds; directly determines peak throughput and top-k width matching.
  • token pruning rate p = 0.1–0.3
    Evaluated at 0.1/0.2/0.3 (DynamicViT); speedup claims (up to 7.8×) depend on these hand-chosen rates.
  • RPU grid size = 2×2 / 1×1
    2×2 on U50, 1×1 on ZCU104; DALO gains (up to 79 %) are measured only on the multi-RPU configuration.
  • top-k max width = 256
    Set to 256; sorter micro-architecture (bitonic vs dual-layer) chosen per device.
assumptions (4)
  • domain assumption All target multimodal layers can be expressed as DDMM, SDDMM or SpMM without material loss of correctness or efficiency.
    Stated in §1 and §3; underpins the single-engine claim.
  • domain assumption Mode-switch overhead (pipeline drain + register update) is negligible relative to kernel run times of thousands of cycles.
    §3.1; required for the 'no reconfiguration' advantage.
  • domain assumption int8 quantization with Bitsandbytes/PyTorch static methods preserves accuracy within 2.5 % for the chosen models.
    §5.3 and Fig. 7; supports the energy/latency claims under the reported datapath.
  • standard math Standard systolic, SIMD and reduction-tree dataflows are correctly realized by the described muxes and PE ALU.
    Background linear-algebra and FPGA PE literature assumed throughout §3.
invented entities (4)
  • Mode-Switchable Engine (MSE)
    purpose: Time-share WS/OS systolic, 1×CS SIMD and RADT on one PE array via interconnect muxes.
    Core architectural contribution; no independent prior silicon evidence outside this paper.
  • Routable Adder Tree (RADT)
    purpose: Handle highly sparse or skewed reductions without a separate sparse engine.
    Named micro-architecture for high-sparsity mode; evidence is only the reported utilization numbers.
  • Dependency-Aware Layer Offloading (DALO)
    purpose: Overlap independent kernels (e.g., Q/K/V) across RPUs using a compile-time DAG.
    Scheduling policy claimed to deliver up to 79 % throughput; no external validation.
  • Width-matched two-stage top-k unit
    purpose: In-stream token pruning matched to array width CS to avoid large global sorters.
    Hardware block enabling the pruning speedups; design is paper-specific.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TRINE: A Token-Aware, Runtime-Adaptive FPGA Inference Engine for Multimodal AI." pith.science (2026). https://pith.science/paper/HRSZPOD5

@misc{pith2026260322867,
  author       = {Pith},
  title        = {Pith review of: TRINE: A Token-Aware, Runtime-Adaptive FPGA Inference Engine for Multimodal AI},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HRSZPOD5}},
  note         = {Machine review of arXiv:2603.22867}
}
read the original abstract

Multimodal stacks that mix ViTs, CNNs, GNNs, and transformer NLP strain embedded platforms because their compute/memory patterns diverge and hard real-time targets leave little slack. TRINE is a single-bitstream FPGA accelerator and compiler that executes end-to-end multimodal inference without reconfiguration. Layers are unified as DDMM/SDDMM/SpMM and mapped to a mode-switchable engine that toggles at runtime among weight/output-stationary systolic, 1xCS SIMD, and a routable adder tree (RADT) on a shared PE array. A width-matched, two-stage top-k unit enables in-stream token pruning, while dependency-aware layer offloading (DALO) overlaps independent kernels across reconfigurable processing units to sustain utilization. Evaluated on Alveo U50 and ZCU104, TRINE reduces latency by up to 22.57x vs. RTX 4090 and 6.86x vs. Jetson Orin Nano at 20-21 W; token pruning alone yields up to 7.8x on ViT-heavy pipelines, and DALO contributes up to 79% throughput improvement. With int8 quantization, accuracy drops remain <2.5% across representative tasks, delivering state-of-the-art latency and energy efficiency for unified vision, language, and graph workloads-in one bitstream.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 4 canonical work pages

  1. [1]

    2024.Configuration Time

    AMD. 2024.Configuration Time. https://docs.amd.com/r/en-US/ug909-vivado- partial-reconfiguration/Configuration-Time Vivado Design Suite User Guide: Dynamic Function eXchange (UG909), Version 2024.2

    2024

  2. [2]

    Jonathan Bachrach, Huy Vo, Brian Richards, Yunsup Lee, Andrew Waterman, Rimas Avižienis, John Wawrzynek, and Krste Asanović. 2012. Chisel: constructing hardware in a Scala embedded language. In49th Annual Design Automation Conference (DAC). ACM, San Francisco, CA, USA, 1216–1225. doi:10.1145/2228360. 2228584

    work page doi:10.1145/2228360 2012

  3. [3]

    Shuning Chang, Pichao Wang, Ming Lin, Fan Wang, David Junhao Zhang, Rong Jin, and Mike Zheng Shou. 2023. Making Vision Transformers Efficient from A Token Sparsification View. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Los Alamitos, CA, USA, 6195–6205. doi:10.1109/ CVPR52729.2023.00600

    arXiv 2023

  4. [4]

    Hanning Chen, Wenjun Huang, Yang Ni, Sanggeon Yun, Yezi Liu, Fei Wen, Alvaro Velasquez, Hugo Latapie, and Mohsen Imani. 2024. TaskCLIP: Extend Large Vision-Language Model for Task Oriented Object Detection. doi:10.48550/ arXiv.2403.08108 arXiv:2403.08108 [cs]

    arXiv 2024

  5. [5]

    Zhiyang Chen, Yousong Zhu, Zhaowen Li, Fan Yang, Chaoyang Zhao, Jinqiao Wang, and Ming Tang. 2024. The Devil is in Details: Delving Into Lite FFN Design for Vision Transformers. InIEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Seoul, Korea, 4130–4134. doi:10.1109/ ICASSP48485.2024.10447756

    arXiv 2024

  6. [6]

    Zhao-Min Chen, Xiu-Shen Wei, Peng Wang, and Yanwen Guo. 2019. Multi-Label Image Recognition with Graph Convolutional Networks. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    2019

  7. [7]

    Mehdi Cherti, Romain Beaumont, Ross Wightman, Mitchell Wortsman, Gabriel Ilharco, Cade Gordon, Christoph Schuhmann, Ludwig Schmidt, and Jenia Jitsev

  8. [8]

    InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Reproducible Scaling Laws for Contrastive Language-Image Learning. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Vancouver, BC, Canada, 2818–2829. doi:10.1109/CVPR52729.2023.00276

    doi:10.1109/cvpr52729.2023.00276 2023

Show all 34 references
  1. [9]

    Peiyan Dong, Mengshu Sun, Alec Lu, Yanyue Xie, Kenneth Liu, Zhenglun Kong, Xin Meng, Zhengang Li, Xue Lin, Zhenman Fang, and Yanzhi Wang. 2023. HeatViT: Hardware-Efficient Adaptive Token Pruning for Vision Transformers. In2023 IEEE International Symposium on High-Performance C...

    2023 doi

  2. [10]

    Qiwei Dong, Xiaoru Xie, and Zhongfeng Wang. 2024. SWAT: An Efficient Swin Transformer Accelerator Based on FPGA. In29th Asia and South Pacific Design Automation Conference (ASP-DAC). 515–520. doi:10.1109/ASP-DAC58780.2024. 10473931

    2024 doi

  3. [11]

    Mohsen Fayyaz, Soroush Abbasi Koohpayegani, Farnoush Rezaei Jafari, Sunando Sengupta, Hamid Reza Vaezi Joze, Eric Sommerlade, Hamed Pirsiavash, and Juergen Gall. 2022. Adaptive Token Sampling for Efficient Vision Transformers. InEuropean Conference on Computer Vision (ECCV). T...

    2022 doi

  4. [12]

    Gerasimos Gerogiannis, Serif Yesil, Damitha Lenadora, Dingyuan Cao, Charith Mendis, and Josep Torrellas. 2023. SPADE: A Flexible and Scalable Accelerator for SpMM and SDDMM. In50th Annual International Symposium on Computer Architecture (ISCA). Orlando, FL, USA, 1–15. doi:10.1...

    2023 doi

  5. [13]

    Rohit Girdhar, Alaaeldin El-Nouby, Zhuang Liu, Mannat Singh, Kalyan Vasudev Alwala, Armand Joulin, and Ishan Misra. 2023. ImageBind: One Embedding Space To Bind Them All. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Vancouver, Bc, Canada

    2023

  6. [14]

    Yuntao Han and Qiang Liu. 2023. HPTA: A High Performance Transformer Ac- celerator Based on FPGA. In33rd International Conference on Field-Programmable Logic and Applications (FPL). 27–33. doi:10.1109/FPL60245.2023.00012

    2023 doi

  7. [15]

    2021.Zynq DPU Product Guide (PG338)

    Xilinx Inc. 2021.Zynq DPU Product Guide (PG338). https://docs.amd.com/r/3.3- English/pg338-dpu Document ID: PG338

    2021

  8. [16]

    Aishwarya Kamath, Mannat Singh, Yann LeCun, Gabriel Synnaeve, Ishan Misra, and Nicolas Carion. 2021. MDETR - Modulated Detection for End-to-End Multi- Modal Understanding. InIEEE/CVF International Conference on Computer Vision (ICCV). Montreal, QC, Canada, 1760–1770. doi:10.11...

    2021 doi

  9. [17]

    Sahar Kazemzadeh, Vicente Ordonez, Mark Matten, and Tamara Berg. 2014. ReferItGame: Referring to Objects in Photographs of Natural Scenes. InConfer- ence on Empirical Methods in Natural Language Processing (EMNLP). Doha, Qatar, 787–798

    2014

  10. [18]

    Zhenglun Kong, Peiyan Dong, Xiaolong Ma, Xin Meng, Wei Niu, Mengshu Sun, Xuan Shen, Geng Yuan, Bin Ren, Hao Tang, Minghai Qin, and Yanzhi Wang

  11. [19]

    In17th European Conference on Computer Vision (ECCV)(Tel Aviv, Israel)

    SPViT: Enabling Faster Vision Transformers via Latency-Aware Soft Token Pruning. In17th European Conference on Computer Vision (ECCV)(Tel Aviv, Israel). 620–640. doi:10.1007/978-3-031-20083-0_37

    doi

  12. [20]

    Bingbing Li, Santosh Pandey, Haowen Fang, Yanjun Lyv, Ji Li, Jieyang Chen, Mimi Xie, Lipeng Wan, Hang Liu, and Caiwen Ding. 2020. FTRANS: energy- efficient acceleration of transformers using FPGA. InACM/IEEE International Symposium on Low Power Electronics and Design (ISLPED)....

    2020 doi

  13. [21]

    Liqiang Lu, Yicheng Jin, Hangrui Bi, Zizhang Luo, Peng Li, Tao Wang, and Yun Liang. 2021. Sanger: A Co-Design Framework for Enabling Sparse Attention using Reconfigurable Architecture. In54th Annual IEEE/ACM International Sym- posium on Microarchitecture (MICRO-54). Virtual, 9...

    2021 doi

  14. [22]

    Dhruv Parikh, Shouyi Li, Bingyi Zhang, Rajgopal Kannan, Carl Busart, and Viktor Prasanna. 2024. Accelerating ViT Inference on FPGA through Static and Dynamic Pruning. InIEEE 32nd Annual International Symposium on Field- Programmable Custom Computing Machines (FCCM). Orlando, F...

    2024 doi

  15. [23]

    Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever

    Alec Radford, Jong Wook Kim, Chris Hallacy, A. Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. 2021. Learning Transferable Visual Models From Natural Language Supervision. In38th Internation...

    2021

  16. [24]

    Yongming Rao, Wenliang Zhao, Benlin Liu, Jiwen Lu, Jie Zhou, and Cho-Jui Hsieh. 2021. DynamicViT: Efficient Vision Transformers with Dynamic Token Sparsification. In35th International Conference on Neural Information Processing Systems (NeurIPS)(Virtual). 13937–13949

    2021

  17. [25]

    Bernstein, Alexander C

    Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael S. Bernstein, Alexander C. Berg, and Li Fei-Fei. 2014. ImageNet Large Scale Visual Recognition Challenge. doi:10.48550/arXiv.1409.0575 arXiv:1...

    2014 doi arXiv

  18. [26]

    Waqas Sultani, Chen Chen, and Mubarak Shah. 2018. Real-world Anomaly Detection in Surveillance Videos. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Salt Lake City, UT, USA, 6479–6488

    2018

  19. [27]

    Shitao Tang, Jiahui Zhang, Siyu Zhu, and Ping Tan. 2022. Quadtree Attention for Vision Transformers. InInternational Conference on Learning Representations (ICLR). Virtual, 1–16

    2022

  20. [28]

    Hanrui Wang, Zhekai Zhang, and Song Han. 2021. SpAtten: Efficient Sparse Attention Architecture with Cascade Token and Head Pruning. InIEEE Inter- national Symposium on High-Performance Computer Architecture (HPCA). Seoul, Korea, 97–110. doi:10.1109/HPCA51647.2021.00018

    2021 doi

  21. [29]

    Kan Wu, Houwen Peng, Zhenghong Zhou, Bin Xiao, Mengchen Liu, Lu Yuan, Hong Xuan, Michael Valenzuela, Xi Stephen Chen, Xinggang Wang, Hongyang Chao, and Han Hu. 2023. TinyCLIP: CLIP Distillation via Affinity Mimicking and Weight Inheritance. InIEEE/CVF International Conference ...

    2023 doi

  22. [30]

    Yifan Xu, Zhijie Zhang, Mengdan Zhang, Kekai Sheng, Ke Li, Weiming Dong, Liqing Zhang, Changsheng Xu, and Xing Sun. 2022. Evo-ViT: Slow-Fast Token Evolution for Dynamic Vision Transformer. In36th AAAI Conference on Artificial Intelligence (AAAI)(Virtual). 2964–2972. doi:10.160...

    2022 doi

  23. [31]

    Haoran You, Zhanyi Sun, Huihong Shi, Zhongzhi Yu, Yang Zhao, Yongan Zhang, Chaojian Li, Baopu Li, and Yingyan Lin. 2023. ViTCoD: Vision Transformer Accel- eration via Dedicated Algorithm and Accelerator Co-Design. InIEEE International Symposium on High-Performance Computer Arc...

    2023 doi

  24. [32]

    Sanggeon Yun, Ryozo Masukawa, Minhyoung Na, and Mohsen Imani. 2024. MissionGNN: Hierarchical Multimodal GNN-based Weakly Supervised Video Anomaly Recognition with Mission-Specific Knowledge Graph Generation. doi:10.48550/arXiv.2406.18815 arXiv:2406.18815 [cs]

    2024 doi arXiv

  25. [33]

    Bingyi Zhang, Rajgopal Kannan, Carl Busart, and Viktor Prasanna. 2024. GCV- Turbo: End-to-end Acceleration of GNN-based Computer Vision Tasks on FPGA. InIEEE 32nd Annual International Symposium on Field-Programmable Custom Computing Machines (FCCM). Orlando, FL, USA, 66–77. do...

    2024 doi

  26. [34]

    Prasanna

    Bingyi Zhang, Rajgopal Kannan, Carl Busart, and Viktor K. Prasanna. 2024. VisionAGILE: A Versatile Domain-Specific Accelerator for Computer Vision Tasks.IEEE Transactions on Parallel and Distributed Systems35, 12 (Dec. 2024), 2405–2422. doi:10.1109/TPDS.2024.3466891

    2024 doi

Pith tools

Reviewed July 13, 2026 · model on record in the stance chip above.