In cooperation with
A curated forum for rigorous, technical discussion at the frontier of AI — accelerating the pace of scientific discovery through deep peer exchange, agentic platforms, and cloud infrastructure.
Inspired by research convenings across frontier labs, Stanford, and Berkeley, we bring together academic researchers, PIs, and engineers building in production to examine concrete work — papers, methods, and empirical results — with a sharp focus on underlying assumptions, evaluation methodology, failure modes, and what counts as convincing evidence.
Each session features 2–3 talks selected for technical rigor and discussion value. Presentations are intentionally brief and focused so the majority of time is reserved for questions, critique, and collaborative debate, with papers shared in advance to ensure a high-baseline conversation.
Designed specifically for Principal Investigators, Lab Leads, and academic researchers, this four-city event series is characterized by the close cooperation between the Frontier Research Club, Google Cloud Partner Digital Schooling, and Google Cloud. While rooted in the open, agnostic spirit of the Frontier Research Club, this series specifically addresses the practical path to cloud-enabled research: lowering the entrance barrier to start experimenting on cloud infrastructure through one-time academic compute grants for faculty and PhD candidates, providing free 48-hour Google Cloud Platform exploration vouchers directly at check-in, and accelerating discovery by embedding frontier foundation models—from AlphaGenome, AlphaEvolve and AlphaFold to Google Antigravity combined with the Agent Development Kit and Agent Runtime—directly into automated scientific workflows.
Hamburg, Munich, Berlin & Zurich at Google Campuses
One-time academic compute grants for faculty & PhDs via cloud.google.com/edu
AlphaGenome, AlphaFold, Gemma & Quantum AI
Exclusive exchange for PIs, Postdocs & PhDs during Apéro
Each session features 2–3 selected talks with intentionally compact presentations so the majority of time is reserved for questions, critique, and technical debate. Papers and supporting materials are shared in advance to ensure a high-baseline conversation focused on assumptions, evaluation methodology, and failure modes.
Have an accepted paper or preprint on autonomous research loops or AI for science? Authors are invited to submit their work for upcoming sessions. Because seating at Google campus locations is strictly limited, admission requires verification via official institutional or university email.
Join us in person across Germany and Switzerland at official Google office locations. Experience high-end conference environments, direct peer exchange, and catering.
Monday, September 21, 2026
check Complimentary Apéro & Partner Exchange
check Limited capacity for academic institutions
Monday, November 9, 2026
check Complimentary Apéro & Partner Exchange
check Deep focus on Autonomous Research Loops
Friday, November 27, 2026
check Complimentary Apéro & Partner Exchange
check Limited capacity for academic institutions
Thursday, December 3, 2026
check Complimentary Apéro & Partner Exchange
check Direct connection to Zurich Main Station (HB)
A compact afternoon of expert keynotes, hands-on implementation, and peer academic exchange.
Welcome reception and badge collection.
Join Alex Schroeder for a 30-minute keynote on hard-won lessons from building and scaling autonomous AI research loops. As frontier systems transition from calculation to independent ideation, scientific discovery throughput is bounded by verification rigor and the Selection-Generation Gap rather than raw model IQ. Explore the full systems blueprint—from stateless deliberation panels and literature grounding to ephemeral GPU cloud execution and cryptographic pre-registration—and discover how verifier-driven architectures overcome agent pathologies like reward hacking and instrument narcissism.
Prof. Pokutta presents his landmark paper The Agentic Researcher, introducing an open-source framework and 5-level taxonomy that turns CLI coding agents into autonomous scientific collaborators. Explore how sandboxed agents run 20+ hour independent multi-node, multi-GPU experiment loops—with rigorous mathematical guardrails, automated code verification, and cluster-scale orchestration.
Prof. Dr. Sören Laue presents his groundbreaking theoretical work uncovering the architectural origins of position bias and the "Lost-in-the-Middle" phenomenon in causal Transformers. Discover how residual-aware attention rollout resolves the open collapse dilemma (Wu et al., 2025) and formally proves that four architectural forces—causal masking, residual connections, relative positional encodings, and content structure—induce broad U-shaped token influence profiles even without semantic content, closely matching empirical measurements on LLMs up to BLOOM-176B.
Networking and refreshment service.
A deep-dive technical walkthrough and live demonstration of Google's Antigravity agentic development ecosystem: Explore how the Antigravity CLI powers headless, sandboxed terminal loops and autonomous research workflows with rigorous verification guardrails. Discover how the Antigravity Desktop App & Platform provides an interactive multi-agent workbench with live artifact inspection and persistent sidecars, and learn how the Google Antigravity SDK enables researchers to author custom subagents, connect Model Context Protocol (MCP) tooling, and orchestrate frontier Gemini models directly across cloud infrastructure.
Accepted as an ICML 2026 Spotlight Paper (within top 2.2% among 23,918 submissions in Seoul), Prof. Dr. Artur Andrzejak and PhD candidate Min Xue (Heidelberg University) present a training-free framework for mechanistic interpretability that operates directly on weight matrices of Transformer MLP sublayers via Singular Value Decomposition (SVD). Discover how their Native Network Anatomy (NaNA) framework decomposes projection matrices into orthogonal Detector-Effector Units (DEUs) and introduces Subspace Contribution Analysis (SCA) to quantify causal contributions to predictions—yielding orders-of-magnitude efficiency gains over trained proxy models like Sparse Autoencoders while preserving complete parameter fidelity.
Google’s AI Co-Scientist serves as a collaborative, multi-agent artificial intelligence system operating within a strict scientist-in-the-loop paradigm (presented live remotely). An orchestrating Supervisor agent coordinates specialized subagents—spanning hypothesis generation, literature critique, simulated debate ranking, evolutionary refinement, and cross-domain semantic proximity—to design novel experimental protocols from natural-language research goals. Explore real-world breakthroughs verified in wet labs, from a novel drug synergy reducing liver fibrosis markers by ~50% in human organoids to repurposing breast cancer therapeutics for acute myeloid leukemia (AML) and deciphering bacterial horizontal gene transfer.
Conclude the day with an informal networking hour to exchange ideas with Google Engineers, Cloud Solutions Architects, and academic peers over drinks and appetizers.
Participation is complimentary for faculty, researchers, lab leads, and academic staff. Please complete the registration form below.