# Sahand Sharifzadeh

> Sahand Sharifzadeh is a Staff Research Scientist on the Gemini team at Google DeepMind in London, working on mid-training, reasoning and multimodal models.

## Key facts
- Sahand Sharifzadeh (سهند شریف زاده) is a Staff Research Scientist on the Gemini Team at Google DeepMind in London, United Kingdom (at Google DeepMind since Aug 2022).
- Developed a key recipe used in Gemini 4 Argon as a core member of the Gemini Mid-Training team; contributed to the Gemini 2 and 3 model series.
- Contributed to Flamingo (NeurIPS 2022) as a DeepMind research scientist intern in 2021; first author and co-lead of Synth² (2024).
- Ph.D. in Computer Science, summa cum laude, LMU Munich (supervisor: Prof. Volker Tresp); M.Sc. TU Munich; B.Sc. Shahid Beheshti University.
- 17,000+ citations on Google Scholar. Languages: Persian, English, Azerbaijani, German, Turkish.
- Also known as: S. Sharifzadeh, S Sharifzadeh, سهند شریف زاده. Handles: @sharifza (GitHub), @sahandsharif (X).
- Research themes: RL & post-training; Multimodal & vision–language; Computer vision; Knowledge graph embeddings; AI for Science; Robotics.

## Profiles
- [Google Scholar](https://scholar.google.com/citations?user=frzfxXYAAAAJ)
- [LinkedIn](https://www.linkedin.com/in/sahandsharifzadeh)
- [GitHub](https://github.com/sharifza)
- [X](https://x.com/sahandsharif)
- [ORCID](https://orcid.org/0000-0002-5534-4017)
- [Website](https://sharifza.github.io/)

## In his own words
Items quoted from the site are in Sahand's own voice: "I" refers to Sahand Sharifzadeh.
Building frontier multimodal and reasoning models across 12+ years of research in foundation models, computer vision, deep RL, and robotics. I developed a key recipe used in Gemini 4 Argon and contributed to the Gemini 2 and 3 model series and Flamingo. Across the Gemini Mid-Training and Thinking areas, I drove research projects on reinforcement learning efficiency and scaling, inference-time scaling, quantization in post-training, and strategic reasoning in games, and co-led research on self-improving VLMs. Earlier, my research spanned looped graph transformers for iterative scene understanding (Classification by Attention, 2020), knowledge graph embeddings (PyKEEN), and deep inverse reinforcement learning for autonomous driving.

## News
- (2026-10) Gemini 4 Argon is now available to everyone, after its announcement on 30 September. I was a core member of the Gemini Mid-Training team and developed a key recipe used in the model. Source: https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/
- (2026-09) Game Arena: strategic LLM evaluation in competitive environments (Kaggle). Technical report on arXiv. Source: https://arxiv.org/abs/2609.31473
- (2025-07) Gemini 2.5 technical report released. Source: https://arxiv.org/abs/2507.06261
- (2024-03) Synth²: self-improving VLMs with synthetic captions and image embeddings (co-lead). Source: https://arxiv.org/abs/2403.07750
- (2024) Ultra-high-granularity detector simulation published in Nature Communications. Source: https://www.nature.com/articles/s41467-024-49104-4

## Selected publications
- [Game Arena: Strategic LLM Evaluation in Competitive Environments](https://arxiv.org/abs/2609.31473): Technical Report · arXiv, 2026
- [Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities](https://arxiv.org/abs/2507.06261): Technical Report, 2025
- [Synth²: Boosting Visual-Language Models with Synthetic Captions and Image Embeddings](https://arxiv.org/abs/2403.07750): arXiv preprint, 2024 (First Author & Co-Lead)
- [Ultra-High-Granularity Detector Simulation with Intra-Event Aware Generative Adversarial Network and Self-Supervised Relational Reasoning](https://www.nature.com/articles/s41467-024-49104-4): Nature Communications, 2024
- [Flamingo: A Visual Language Model for Few-Shot Learning](https://arxiv.org/abs/2204.14198): NeurIPS, 2022 (10,000+ Citations)
- [Relationformer: A Unified Framework for Image-to-Graph Generation](https://arxiv.org/abs/2203.10202): ECCV, 2022
- [Bringing Light Into the Dark: A Large-scale Evaluation of Knowledge Graph Embedding Models Under a Unified Framework](https://arxiv.org/abs/2006.13365): IEEE TPAMI, 2021
- [PyKEEN 1.0: A Python Library for Training and Evaluating Knowledge Graph Embeddings](https://jmlr.org/papers/v22/20-825.html): JMLR, 2021
- [Automatic Image-Based Determination of Pruning Mass as a Determinant for Yield Potential in Grapevine Management and Breeding](https://doi.org/10.1111/ajgw.12243): Australian Journal of Grape and Wine Research, 2017 (Stereo 3D Reconstruction)
- [Learning to Drive using Inverse Reinforcement Learning and Deep Q-Networks](https://arxiv.org/abs/1612.03653): NeurIPS Workshop on Deep Learning for Action and Interaction, 2016 (First Author)

## Docs
- [Full profile](https://sharifza.github.io/llms-full.txt): experience, research trajectory, talks, all 26 publications and awards
- [Homepage](https://sharifza.github.io/)

Last updated: 2026-10-24