Reading group on Sustainable Machine Learning and ICT. We discuss papers every other week, usually on Wednesdays at 3.30PM CET. If interested reach out to Dustin or Raghav.

1. Upcoming Meetings
2. Past Meetings
3. About
4. Contact

• Raghavendra Selvan will present Mamba: Linear-Time Sequence Modeling with Selective State Spaces (Gu and Dao 2023).
• Paper

• Jonathan Wenshøj will present Gradient ℓ1 Regularization for Quantization Robustness (Alizadeh et al. 2020).
• Paper

• Frederik L Johansen presented QLoRA: Efficient Finetuning of Quantized LLMs (Dettmers et al. 2023)
• Paper

• Sebastian Hammer Eliassen presented Activation Compression of Graph Neural Networks using Block-wise Quantization with Improved Variance Minimization (Eliassen et al. 2024)
• Paper

• Bob Pepin presented How Does Information Bottleneck Help Deep Learning?
• Paper

• Sebastian Hammer Eliassen presented BitGNN: Unleashing the Performance Potential of Binary Graph Neural Networks on GPUs (Chen et al. 2023)
• Paper

• Eya Ben Chaaben presented Towards rapid interactive machine learning: evaluating tradeoffs of classification without representation. (Arendt et al. 2019)
• Paper

• Daniel Geißler presented Read the Signs: Towards Invariance to Gradient Descent's Hyperparameter Initialization (Wadi et al. 2023)
• Paper

• Pedram Bakhtiarifard presented Neural Architecture Search without Training (Mellor et al. 2020)
• Paper

• Dustin Wright will be presented Language Modeling is Compression (Delétang et al. 2023).
• Paper

• Raghavendra Selvan presented Pruning vs Quantization: Which is Better? (Kuzmin et al. 2023).
• Paper

• Tong Chen presented Sparse Polynomial Optimization and Its Application to Deep Neural Network.
• Paper

• Pedram Bakhtiarifard presented Once-for-All: Train One Network and Specialize it for Efficient Deployment (Cai et al. 2020)
• Paper

• Eya Ben Chaaben presented EnergyVis: Interactively tracking and exploring energy consumption for ML models (Shaikh et al. 2021)
• Paper

• Daniel Geißler presented Dodging the Sparse Double Descent (Quétu and Tartaglione 2023)
• Paper

• Sebastian Hammer Eliassen will present EXACT: Scalable Graph Neural Networks Training via Extreme Activation Compression (Liu et al. 2022)
• Paper

• Emil Jørgensen Njor presented Data Aware Neural Architecture Search (Njor et al. 2023)
• Paper

• Julian Schön presented Learning from Randomly Initialized Neural Network Features (Amid et al. 2022)
• Paper

• Bo Zhou presented Overcoming Oscillations in Quantization-Aware Training (Nagel et al. 2022)
• Paper

• Raghavendra Selvan presented 8-bit Optimizers via Block-wise Quantization (Dettmers et al. 2022)
• Paper

• Dustin Wright presented An Information-Theoretic Justification for Model Pruning (Isik et al, 2022)
• Paper

The reading group is part of the dissemination activities related to two EU projects Raghav is involved with. European Union’s Horizon Europe Research and Innovation programme under grant agreements No. 101070284 (EnrichMyData) and No. 101070408 (SustainML).

For more details about the project, reading group, or for opportunities reach out to Raghavendra Selvan: [email protected].