Song Han
Associate Professor
Develops efficient deep learning, including model compression, quantization, and hardware-software co-design for LLM inference.
Song Han's lab at MIT works on efficient training & inference and accelerators, chips & architecture. 52 papers since 2023, 46 at top venues, mostly ICLR and ICML. 7 PhD students. 13 alumni, most often to NVIDIA and Meta. Applications close .
At a glance
- papers, 2023–2026
- 14/ year
- at T1 venues
- 88%
- PhD students
- 7
- MS students
- 0
- alumni listed
- 13
Papers by yearfrom the lab homepage, 2023–2026
- 13 papers2026 · T1 10 · collecting
- 19 papers2025 · T1 18
- 11 papers2024 · T1 11
- 9 papers2023 · T1 7
T1 venuesOtherVenues: ICLR, ICML, CVPR, ICCV, MLSys
Research focuspaper titles that use each topic's words
- Efficient training & inference29 papers
- Accelerators, chips & architecture1 paper
- develops efficient0 papers
- hardware-software0 papers
- hardware-software co-design0 papers
Paper list
From the lab homepage.
202613 papers
- CVPR (T1)DeltaQuant: 4-bit Video Diffusion Models with Spatiotemporal Delta Smoothing
- ICLR (T1)Fast-dLLM v2: Efficient Block-Diffusion Large Language Model
- ICLR (T1)Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding
- CVPR HighlightForeAct: Steering Your VLA with Efficient Visual Foresight Planning
- ICML (T1)FourTune: Towards Fully 4-Bit Efficient Post-Training for Diffusion Models
- ICLR OralLocality-aware Parallel Decoding for Efficient Autoregressive Image Generation
- ICLR (T1)LongLive: Real-time Interactive Long Video Generation
- ICLR (T1)QeRL: Beyond Efficiency -- Quantization-enhanced Reinforcement Learning for LLMs
- ICML (T1)Quant VideoGen: Auto-Regressive Long Video Generation via 2-Bit KV-Cache Quantization
- ICLR OralSANA-Video: Efficient Video Generation with Block Linear Diffusion Transformer
- ICLR (T1)StreamingVLM: Real-Time Understanding for Infinite Video Streams
- ASPLOS (T1)Taming the Long-Tail: Efficient Reasoning RL Training with Adaptive Drafter
202519 papers
- ICLR (T1)COAT: Compressing Optimizer states and Activation for Memory-Efficient FP8 Training
- ICCV (T1)DC-AE 1.5: Accelerating Diffusion Model Convergence with Structured Latent Space
- ICCV (T1)DC-AR: Efficient Masked Autoregressive Image Generation with Deep Compression Hybrid Tokenizer
- ICLR (T1)Deep Compression Autoencoder for Efficient High-Resolution Diffusion Models
- ICLR (T1)DuoAttention: Efficient Long-Context LLM Inference with Retrieval and Streaming Heads
- ICLR (T1)HART: Efficient Visual Generation with Hybrid Autoregressive Transformer
- —HPCA 2025
- NeurIPS (T1)Jet-Nemotron: Efficient Language Model with Post Neural Architecture Search
- NeurIPS (T1)Long-RL: Scaling RL to Long Videos
- ICLR (T1)LongVILA: Scaling Long-Context Visual Language Models for Long Videos
- MLSys (T1)LServe: Efficient Long-sequence LLM Serving with Unified Sparse Attention
- MLSys (T1)QServe: W4A8KV4 Quantization and System Co-design for Efficient LLM Serving
202411 papers
- ISCA (T1)Atomique: A Quantum Compiler for Reconfigurable Neutral Atom Arrays
- MLSys (T1)AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration
- CVPR (T1)Condition-Aware Neural Network for Controlled Image Generation
- CVPR (T1)DistriFusion: Distributed Parallel Inference for High-Resolution Diffusion Models
- ICLR (T1)Efficient Streaming Language Models with Attention Sinks
- IJCV (T1)FastComposer: Tuning-Free Multi-Subject Image Generation with Localized Attention
- ICLR (T1)LongLoRA: Efficient Fine-tuning of Long-Context Large Language Models
- DAC (T1)Q-Pilot: Field Programmable Qubit Array Compilation with Flying Ancillas
- ICML (T1)Quest: Query-Aware Sparsity for Efficient Long-Context LLM Inference
- ECCV (T1)Sparse Refinement for Efficient High-Resolution Semantic Segmentation
- CVPR (T1)VILA: On Pre-training for Visual Language Models
20239 papers
- ICRA (T1)BEVFusion: Multi-Task Multi-Sensor Fusion with Unified Bird's-Eye View Representation
- ICCV (T1)EfficientViT: Multi-Scale Linear Attention for High-Resolution Dense Prediction
- CVPR (T1)FlatFormer: Flattened Window Attention for Efficient Point Cloud Transformer
- MICRO (T1)PockEngine: Sparse and Efficient Fine-tuning in a Pocket
- ISCA (T1)Retrospective: EIE: Efficient Inference Engine on Sparse and Compressed Neural Network
- ICML (T1)SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models
- ISLPEDSpAtten-Chip: A Fully-Integrated Energy-Scalable Transformer Accelerator Supporting Adaptive Model Configuration and Word Elimination for Language Understanding on Edge Devices
- NeurIPS / / , MICRO , ICML , MLSys , IEEE CAS MagazineTiny Machine Learning Projects
- MICRO (T1)TorchSparse++: Efficient Training and Inference Framework for Sparse Convolution on GPUs
Where alumni went13 listed, 40% to industry
Industry 4Academia 6
Most often to
- NVIDIA
- Meta
- OpenAI
Also to
- MIT
- NSF AI Institute
- the NSF Athena AI Institute
Apply to MITFall 2027
MIT EECS Computer Science PhD
- Round
- Single round
- Opens
- —
- GRE
- See program page
- English
- TOEFL/IELTS
- Fee
- USD 90
Confirmed on the program page.
Similar labsnearest by research summary
- SGNUS
Yao Lu
- Distributed, cloud & edge systems
- USUIUC
Yongjoo Park
- Efficient training & inference
- USUCSD
Farinaz Koushanfar
- Efficient training & inference
- USUT Austin
Diana Marculescu
- Accelerators, chips & architecture
- USGeorgia Tech
Yingyan Celine Lin
- Accelerators, chips & architecture
- SGNUS
Wenqi Jiang
- 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.
MITcloses