SqueezeAILab/open_source_projects

Open Source Projects from Pallas Lab

Open Source Projects from PALLAS Lab

Below are links to different open source projects from Prof. Keutzer's lab at UC Berkeley.

Core Optimization Algorithms

• AdaHessian: A Second-order Optimization Algorithm
• HessianFlow: A Library for Hessian Based Algorithms in Machine Learning
• PowerNorm: Rethinking Batch Normalization in Transformers
• PyHessian: Neural Networks Through the Lens of the Hessian
• TRAttack: Trust Region Adversarial Attack

Neural Network Architecture Design

• ANODE: Unconditionally Accurate Memory-efficient Gradients for Neural ODEs
• DiracDeltaNet
• Image2Point
• LiDAR
• SqueezeNext: Hardware-aware Neural Network Design
• ShiftNet
• SqueezeDet
• SqueezeSeg
• SqueezeSegV2
• SqueezeSegV3
• YOGO

Efficient Inference and Compression

• BitPack: A Practical Tool to Efficiently Save Ultra-Low Precision/Mixed-Precision Quantized Models
• CoDeNet: Efficient Deployment of Input-Adaptive Object Detection on Embedded FPGAs
• HAP: Hessian Aware Pruning and Neural Implant
• HAWQ: Hessian Aware Quantization
• I-BERT: Integer Only BERT Quantization
• Q-ASR: Integer-only Zero-shot Quantization for Efficient Speech Recognition
• ZeroQ: A Novel Zero Shot Quantization Framework
• LTP: Learned Token Pruning

Domain Randomization and Domain Adaptation

• MADAN
• PCS for Few-shot Unsupervised Domain Adaptation