Pub
publications by categories in reversed chronological order. generated by jekyll-scholar.
2023
- Neural network activation compression with non-uniform mantissasJan 2023US Patent 11,562,247
- Architectural backdoors in neural networksIn Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Jan 2023
- Task-Agnostic Graph Neural Network Evaluation via Adversarial CollaborationarXiv preprint arXiv:2301.11517, Jan 2023
- Dynamic Stashing Quantization for Efficient Transformer TrainingarXiv preprint arXiv:2303.05295, Jan 2023
- Revisiting Automated Prompting: Are We Actually Doing Better?arXiv preprint arXiv:2304.03609, Jan 2023
- NEURAL NETWORK ACTIVATION COMPRESSION WITH NON-UNIFORM MANTISSASMay 2023US Patent App. 18/092,876
- Adaptive Channel Sparsity for Federated Learning Under System HeterogeneityIn Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, May 2023
- Hybrid Graph: A Unified Graph Representation with Datasets and Benchmarks for Complex GraphsarXiv preprint arXiv:2306.05108, May 2023
- Genomic Interpreter: A Hierarchical Genomic Deep Neural Network with 1D Shifted Window TransformerarXiv preprint arXiv:2306.05143, May 2023
- The Curse of Recursion: Training on Generated Data Makes Models ForgetarXiv preprint arxiv:2305.17493, May 2023
- Fast Prototyping Next-Generation Accelerators for New ML Models using MASE: ML Accelerator System ExplorationarXiv preprint arXiv:2307.15517, May 2023
- Will More Expressive Graph Neural Networks do Better on Generative Tasks?arXiv preprint arXiv:2308.11978, May 2023
2022
- DAdaQuant: Doubly-adaptive quantization for communication-efficient Federated LearningIn International Conference on Machine Learning, May 2022
- Model Architecture Adaption for Bayesian Neural NetworksarXiv preprint arXiv:2202.04392, May 2022
- Efficient Adversarial Training With Data PruningarXiv preprint arXiv:2207.00694, May 2022
- Software and Hardware Co-design for Efficient Neural NetworksUniversity of Cambridge, May 2022
- Augmentation BackdoorsarXiv preprint arXiv:2209.15139, May 2022
- ImpNet: Imperceptible and blackbox-undetectable backdoors in compiled neural networksarXiv preprint arXiv:2210.00108, May 2022
- Wide Attention Is The Way Forward For TransformersarXiv preprint arXiv:2210.00640, May 2022
- DARTFormer: Finding The Best Type Of AttentionarXiv preprint arXiv:2210.00641, May 2022
- Revisiting Structured DropoutarXiv preprint arXiv:2210.02570, May 2022
- Revisiting Embeddings for Graph Neural NetworksIn Learning on Graphs Conference, May 2022
- Flareon: Stealthy any2any Backdoor Injection via Poisoned AugmentationarXiv preprint arXiv:2212.09979, May 2022
- Rapid Model Architecture Adaption for Meta-LearningIn Thirty-sixth Conference on Neural Information Processing Systems (NeurIPS), May 2022
2021
- Sponge examples: Energy-latency attacks on neural networksIn 2021 IEEE European Symposium on Security and Privacy (EuroS&P), May 2021
- Manipulating sgd with data ordering attacksAdvances in Neural Information Processing Systems, May 2021
- Markpainting: Adversarial Machine Learning meets InpaintingIn 38th International Conference on Machine Learning, May 2021
- FedDrop: Trajectory-weighted Dropout for Efficient Federated LearningMay 2021
2020
- Blackbox attacks on reinforcement learning agents using approximated temporal informationIn 2020 50th Annual IEEE/IFIP International Conference on Dependable Systems and Networks Workshops (DSN-W), May 2020
- Towards certifiable adversarial sample detectionIn Proceedings of the 13th ACM Workshop on Artificial Intelligence and Security, May 2020
- Pay Attention to Features, Transfer Learn Faster CNNsIn International Conference on Learning Representations, May 2020
- Probabilistic Dual Network Architecture Search on GraphsarXiv preprint arXiv:2003.09676, May 2020
- Learned Low Precision Graph Neural NetworksarXiv preprint arXiv:2009.09232, May 2020
- Nudge Attacks on Point-Cloud DNNsarXiv preprint arXiv:2011.11637, May 2020
- Adjusting activation compression for neural network trainingAug 2020US Patent App. 16/276,395
- Neural network activation compression with narrow block floating-pointJul 2020US Patent App. 16/237,197
- Neural network activation compression with outlier block floating-pointJul 2020US Patent App. 16/237,202
2019
- Dynamic Channel Pruning: Feature Boosting and SuppressionIn International Conference on Learning Representations (ICLR), Jul 2019
- Sitatapatra: Blocking the Transfer of Adversarial SamplesarXiv preprint arXiv:1901.08121, Jul 2019
- Focused quantization for sparse CNNsIn Advances in Neural Information Processing Systems, Jul 2019
- Characterizing Sources of Ineffectual Computations in Deep Learning NetworksIn 2019 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS), Jul 2019
- Automatic generation of multi-precision multi-arithmetic CNN accelerators for FPGAsIn 2019 International Conference on Field-Programmable Technology (ICFPT), Jul 2019
2018
- Redundancy-reduced mobilenet acceleration on reconfigurable logic for imagenet classificationIn Applied Reconfigurable Computing. Architectures, Tools, and Applications: 14th International Symposium, ARC 2018, Santorini, Greece, May 2-4, 2018, Proceedings 14, Jul 2018
- Mayo: A Framework for Auto-generating Hardware Friendly Deep Neural NetworksIn Proceedings of the 2nd International Workshop on Embedded and Mobile Deep Learning (EMDL 2018), Jul 2018
- To compress or not to compress: Understanding the Interactions between Adversarial Attacks and Neural Network CompressionIn The Conference on Systems and Machine Learning (SysML), Jul 2018
- The Taboo Trap: Behavioural Detection of Adversarial SamplesarXiv preprint arXiv:1811.07375, Jul 2018
2016
- An efficient implementation of online arithmeticIn 2016 International Conference on Field-Programmable Technology (FPT), Jul 2016