Scholar

Ramé Alexandre

Google Scholar ID: 7znwivwAAAAJ

Google DeepMind

Deep LearningGeneralizationAlignment

Citations & Impact

All-time

Citations

4,016

H-index

i10-index

Publications

Co-authors

list available

Contact

Publications

20 items

Browse publications on Google Scholar (top-right) ↗

Resume (English only)

Academic Achievements

PhD thesis awarded the 2024 SSFAM prize for best French PhD in Machine Learning
Contributed to Gemma 3 Technical Report (arXiv), introducing a novel post-training recipe that significantly improves math and chat capabilities
Proposed Diversity-Rewarded CFG Distillation (ICLR 2025), a fine-tuning method combining RL and weight averaging to enhance distillation
Introduced WARP (arXiv), leveraging weight-averaged policies to improve the KL-reward trade-off in RLHF
Proposed WARM (ICML 2024), a reward modeling strategy that merges multiple reward models for improved robustness and reduced reward hacking
PhD research (2023) analyzed how weight-averaged ensembling improves out-of-distribution generalization and alignment
Co-authored ICLR 2024 paper on hallucinations and explainability in large multimodal models

Co-authors

6 total