Scholar
N. Benjamin Erichson
Google Scholar ID: 8ViYcioAAAAJ
Research Scientist
Citations & Impact
Citations
2,764
H-index
24
i10-index
40
Publications
20
Co-authors
21
list available
Publications
Resume (English only)
Academic Achievements
- Multiple papers accepted at top-tier conferences: NeurIPS 2025, ICML 2025, ICLR 2025 (including spotlight), AISTATS 2025, ICLR 2024 (including spotlight), AISTATS 2024, AISTATS 2023 (oral), ICLR 2022 (including spotlight).
- Notable papers include 'Block-Biased Mamba for Long-Range Sequence Processing', 'Emoji Attack: Enhancing Jailbreak Attacks Against Judge LLM Detection', 'Tuning Frequency Bias of State Space Models', and 'HOPE for a Robust Parameterization of Long-memory State Space Models'.
- Serving as Area Chair for NeurIPS 2025, ICML 2025, and ICLR 2025.
- Co-organizing the Deep Learning for Science Summer School.
- Co-organizing the Berkeley Lab AI for Science Summit (BLASS 24).
Background
- Currently a Research Scientist at Lawrence Berkeley National Laboratory.
- Leads the Deep Learning Group at the International Computer Science Institute (ICSI), an affiliated institute of UC Berkeley.
- Broadly interested in understanding how deep learning systems work and improving their robustness, interpretability, and efficiency.
- Applies a scientific approach to neural networks, using dynamical systems theory to explain issues like vanishing/exploding gradients.
- Currently working on large-scale generative diffusion models for spatio-temporal forecasting in earth science and fluid dynamics.
- Exploring how foundation models can integrate reasoning and multimodal information for better predictions.
- Increasing focus on AI safety, particularly understanding and mitigating vulnerabilities in large language models such as jailbreaking and backdoor attacks.
