Guohao Dai
副教授
Works on sparse computing circuits and systems for efficient AI inference.
Guohao Dai's lab at SJTU works on efficient training & inference and accelerators, chips & architecture. 78 papers since 2023, 23 at top venues, mostly ICML and MLSys. 11 PhD students and 1 master's student. 2 alumni, most often to Shenzhen Yinwang Intelligent Technologies Co.. Applications close .
At a glance
- papers, 2023–2026
- 20/ year
- at T1 venues
- 29%
- PhD students
- 11
- MS students
- 1
- alumni listed
- 2
Papers by yearfrom the lab homepage, 2023–2026
- 13 papers2026 · T1 6 · collecting
- 23 papers2025 · T1 8
- 19 papers2024 · T1 4
- 23 papers2023 · T1 5
T1 venuesOtherVenues: ICML, MLSys, AAAI, CVPR, DAC
Research focuspaper titles that use each topic's words
- Efficient training & inference32 papers
- efficient24 papers
- Accelerators, chips & architecture16 papers
- sparse11 papers
- circuits0 papers
Paper list
From the lab homepage.
202613 papers
- Asia and South Pacific Design Automation ConferenceBalanceGS: Algorithm-System Co-design for Efficient 3D Gaussian Splatting Training on GPU
- IEEE Transactions on Computer-Aided Design of Integrated Circuits and SystemsDiffAttn: Sparse Attention Architecture for Video Generation with Spatial Information-guided Differential Estimation
- IEEE European Solid-State Electronics Research Conference (ESSERC), IEEEDiFlash: a 13.70TOPS/W 156.05Pixels/mJ Video Generation Accelerator with Bubble-Free Online-Quantized FlashAttention and Query-Reuse Sliding-Tile Attention
- EuroSys (T1)Efficient and Adaptable Overlapping for Computation and Communication via Signaling and Reordering
- Design, Automation and Test in Europe Conference (DATE), IEEEEndor: Exploit Nearly-Decode-Only Opportunities of LLM Reasoning on Near-Memory Architecture
- DATEFAST: A Scalable Framework for Accelerating Flexible Structured Sparse Training
- AAAI (T1)MoSs: Mixture of Scales for Efficient High-Resolution Autoregressive Image Generation
- Asia and South Pacific Design Automation ConferenceSpAct-NDP: Efficient LLM Inference via Sparse Activation on NDP-GPU Heterogeneous Architecture
- AAAI (T1)SpecDiff: Accelerating Diffusion Model Inference with Self-Speculation
- International Conference on Architectural Support for Programming Languages and Operating Systems, ACMSpeContext: Enabling Efficient Long-context Reasoning with Speculative Context Sparsity in LLMs
- ICML (T1)SpecPrune-VLA: Accelerating Vision-Language-Action Models via Action-Aware Self-Speculative Pruning
- ICML (T1)STEP: Warm-Started Visuomotor Policies with Spatiotemporal Consistency Prediction
202523 papers
- DAC (T1)A Cross-model Fusion-aware Framework for Optimizing (gather-matmul-scatter)s Workload
- ASP-DACAccelerator for LLM-Enhanced GNN with Product Quantization and Unified Indexing
- DATEAiSpGEMM: Accelerating Imbalanced SpGEMM on FPGAs with flexible interconnect and intra-row parallel merging
- Transactions on Reconfigurable Technology and Systems, ACMCD-LLM: A Heterogeneous Multi-FPGA System for Batched Decoding of 70B+ LLMs using a Compute-Dedicated Architecture
- —DATIS: DRAM Architecture and Technology Integrated Simulation
- ASP-DACDeploying Diffusion Models with Latency-Oriented Scheduling and Memory Overflow Prevention Based on Graph Optimization
- ACM MMDlfr-vae: Dynamic latent frame rate vae for video generation
- DATEDyLGNN: Efficient LM-GNN Fine-tuning with Dynamic Node Partitioning, Low-degree Sparsity, and Asynchronous Sub-batch
- Proceedings of the Conference on Empirical Methods in Natural Language Processing: Tutorial AbstractsEfficient Inference for Large Language Models–Algorithm, Model, and System
- IEEE Transactions on Computers, IEEEFlashDecoding++Next: High Throughput LLM Inference with Latency and Memory Optimization
- ACM/SIGDA International Symposium on Field-Programmable Gate Arrays (FPGA), ACMFlightVGM: Efficient Video Generation Model Inference with Online Sparsification and Hybrid Precision on FPGAs
- DAC (T1)Harnessing Conventional Video Processing Insights for Emerging 3D Video Generation Models: A Comprehensive Attention-aware Way
202419 papers
- TCADA Point Transformer Accelerator with Distribution-aware Heuristic Distance Calculation
- DATEDyPIM: Dynamic-inference-enabled Processing-In-Memory Accelerator
- Design Automation Conference (DAC), IEEE/ACMDySpMM: From Fix to Dynamic for Sparse Matrix-Matrix Multiplication Accelerators
- SOCCEfficient Deployment of Large Language Model across Cloud-Device Systems
- IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems (TCAD), 2226-2239, 44,6Enabling Efficient Sparse Multiplications on GPUs with Heuristic Adaptability
- ICML (T1)Evaluating Quantized Large Language Models
- International Conference on Computer-Aided Design (ICCAD), IEEE/ACMFast and Efficient 2-bit LLM Inference on GPU: 2/4/16-bit in a Weight Matrix with Asynchronous Dequantization
- International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS), 349-366, ACMFEASTA: A Flexible and Efficient Accelerator for Sparse Tensor Algebra in Machine Learning
- Transactions on Computer-Aided Design of Integrated Circuits and Systems, IEEEFine-Grained Structured Sparse Computing for FPGA-Based AI Inference
- MLSys (T1)FlashDecoding++: Faster Large Language Model Inference with Asynchronization, Flat GEMM Optimization, and Heuristics
- CVPR (T1)FlashEval: Towards Fast and Accurate Evaluation of Text-to-image Diffusion Generative Models
- FPGAFlightLLM: Efficient Large Language Model Inference with a Complete Mapping Flow on FPGAs
202323 papers
- International Conference on Computer- Aided Design（ICCAD）, IEEE/ACMA Point Transformer Accelerator with Fine-Grained Pipelines and Distribution-Aware Dynamic FPS
- ICCV (T1)Ada3D: Exploiting the Spatial Redundancy with Adaptive Inference for Efficient 3D Object Detection
- IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems（TCAD）Adaptive Multi-Dimensional Parallel Fault Simulation Framework on Heterogeneous System
- Design Automation Conference (DAC), IEEE/ACMAn Efficient Accelerator for Point-based and Voxel-based Point Cloud Neural Networks
- Design, Automation and Test in Europe Conference (DATE), IEEECLAP: Locality Aware and Parallel Triangle Counting with Content Addressable Memory
- WWW (T1)CogDL: A Comprehensive Library for Graph Deep Learning
- TCADCoGNN: An Algorithm-Hardware Co-Design Approach to Accelerate GNN Inference with Mini-Batch Sampling
- IEEE/ACM International Symposium on Microarchitecture（MICRO）DF-GAS: a Distributed FPGA-as-a-Service Architecture towards Billion-Scale Graph- based Approximate Nearest Neighbor Search
- MLSys (T1)Exploiting Hardware Utilization and Adaptive Dataflow for Efficient Sparse Convolution in 3D Point Clouds
- TCADGibbon: An Efficient Co-Exploration Framework of NN model and Processing-In-Memory Architecture
- Multimedia Asia Conference (MMAsia), ACMHistory-Detr: Optimize Query Initialization Strategy by Using Historical Information and Kinematics
- MLSys (T1)HyperGef: A Framework Enabling Efficient Fusion for Hypergraph Neural Network on GPUs
Where alumni went2 listed, 100% to industry
Industry 1Academia 0
Most often to
- Shenzhen Yinwang Intelligent Technologies Co.
Apply to SJTUFall 2027
Computer Science PhD (international applicants)
- Round
- Self-funded
- Opens
- —
- GRE
- See program page
- English
- See program page
- Fee
- See program page
Not yet confirmed for this intake.
Similar labsnearest by research summary
- GBOxford
Hamed Haddadi
- Efficient training & inference
- CNSJTU
Xiaofeng Hou
- Accelerators, chips & architecture
- CNSJTU
Yu Feng
- Efficient training & inference
- USGeorgia Tech
Alexey Tumanov
- Efficient training & inference
- CNSJTU
Jieru Zhao
22 T1 - GBCambridge
Robert Mullins
- Accelerators, chips & architecture
Your notes
Save this lab to keep notes.
Data
Checked against the lab homepage.
- Collected
- Verified
- Papers
- The lab homepage
- Members
- Lab homepage
- Alumni
- Lab homepage
Tiers are venue classifications, not lab rankings.
SJTUcloses