Ever wonder if a computer can simulate brain activity? Google Research just curated a new #NotebookLM notebook asking just this. It features sources on: ✅ Predicting neural activity with AI ✅ Nanoscale brain mapping ✅ Synapse-level reconstructions Start exploring the frontier of neuroscience today: https://lnkd.in/gjyGr4Br
About us
From conducting fundamental research to influencing product development, our research teams have the opportunity to impact technology used by billions of people every day. We aspire to make discoveries that impact everyone, and sharing our research and tools to fuel progress in the field is fundamental to our approach.
- Website
-
https://research.google/
External link for Google Research
- Industry
- Technology, Information and Internet
- Company size
- 1,001-5,000 employees
Updates
-
Measles outbreaks are on the rise, but gaps in data can make it difficult for public health workers to reach vulnerable populations. To help stop further spread of the disease, researchers at Harvard, Mount Sinai and Boston Children's Research used Google's Earth AI to fill data gaps and estimate vaccination rates with "superresolution" — as deep as the zip code level — so they could find hotspots of undervaccination that correlate to measles outbreaks. More in the Nature Health paper → https://lnkd.in/gyvvXrrK
-
-
A critical gap in modern AI isn't language or vision. It's spatial grammar. And it reveals a fundamental data bottleneck. We built MapTrace, a fully automated, generative AI pipeline (models act as creators/critics) to generate 2M high-quality map-path pairs. The result: Fine-tuning on this synthetic data lowered path-tracing errors and boosted the success rate by +6.4 points for Gemini 2.5 Flash on real-world maps. We've open-sourced the 2M question/answer pairs dataset for the research community to build the next generation of intuitive navigation and robotics. Check it out: goo.gle/4tLUza8
-
-
Once upon a time, connections to cryptography suggested that efficiently learning halfspaces with adversarial label noise might be impossible (in the distribution-free setting). Yishay Mansour & co-authors showed otherwise in their 2005 paper, “Agnostically Learning Half Spacesâ€, which won an IEEE Computer Society FOCS Test of Time Award — congratulations!
-
A core fallacy of algorithmic job scheduling is that computing resources are static. They are not. In modern cloud infrastructure, capacity is time-varying. High-priority tasks leave lower-priority, non-preemptive batch jobs vulnerable. An interruption means complete loss of progress. Our new work provides the first constant-competitive algorithm to maximize throughput in this volatile environment, offering the theoretical foundation for robust cloud schedulers.
-
-
Google Research reposted this
We are accelerating mathematical and scientific discovery with AI. Specifically, using advanced AI models based on ✨ Gemini Deep Think and its advanced variants. Two new papers published by Google DeepMind and Google Research highlight a cross-disciplinary effort using Gemini Deep Think to solve research problems across mathematics, physics, and computer science. ✨ For research-level math, a math research agent powered by Gemini Deep Think has already enabled several advancements, produced via varying levels of autonomous research: â–¶ï¸ Reliable autonomous research which calculates certain structure constants in arithmetic geometry called eigenweights, â–¶ï¸ AI-guided collaboration, proving bounds on systems of interacting particles called independent sets, andÂ â–¶ï¸ Extensive semi-autonomous evaluation of 700 open problems on Bloom’s ErdÅ‘s Conjectures database, including autonomous solutions to four open questions listed there. Expanding to Physics and Computer Science, key highlights include: ✨ Refuting Long-standing Conjectures. In computer science, successfully refuted a decade-old conjecture in online submodular optimization. By engineering a specific three-item combinatorial counterexample, it rigorously disproved intuition that had challenged experts since 2015. ✨ Connecting Disparate Fields. Breaking deadlocks in classic algorithmic puzzles, such as "Max-Cut" and "Steiner Tree". It achieves this by "thinking outside the box"—pulling advanced tools from unrelated branches of continuous mathematics, like the Kirszbraun Theorem, to solve discrete computer science problems. ✨ Vibe-Proving & The Advisor Model. A powerful "Advisor" model for collaboration, which involves human-led "Vibe-Proving" cycles where researchers guide the AI through iterative loops to validate intuition, hunt for counterexamples, and refine complex proofs. ✨ Solving Bottlenecks in Physics. In the study of cosmic strings, Gemini identified a novel analytical solution using Gegenbauer polynomials. This allowed researchers to collapse an infinite series into a closed-form finite sum, resolving a technical bottleneck in calculating gravitational radiation. AI played a pivotal role in advancing state-of-the-art research. Some of the recurring patterns and effective strategies include agentic execution loops, deep technical review and bug detection, deep literature synthesis and connection, counter example generation, algorithmic insight and optimization, automated proof generation and verification, interactive refinement, and theoretical justification of heuristics. AI as an amplifier of human ingenuity. Read more in the blog by Thang Luong and Vahab Mirrokni: https://lnkd.in/d-95v9BQ Research papers: Towards autonomous mathematics research: https://lnkd.in/ddiYR6Vz Accelerating Scientific Research with Gemini https://lnkd.in/dsUqnNGS
-
-
Google Research reposted this
How could AI act as a better research collaborator? 🧑🔬 In two new papers with Google Research, we show how Gemini Deep Think uses agentic workflows to help solve research-level problems in mathematics, physics, and computer science. Find out more → https://goo.gle/4aGs3Pz
-
-
For truly realistic conversational research, we must rethink fully autonomous agent design. DialogLab, our new open-source prototyping framework, uses a human-in-the-loop control strategy to achieve realistic human-AI group simulation, offering a necessary alternative to fully autonomous agents. Evaluations with domain experts found that its "Human Control" mode (where you can edit, accept, or dismiss real-time AI suggestions) was preferred in realism, effectiveness, & engagement. DialogLab transforms dialogue design from rigid scripts to spontaneous, adaptable group dynamics. Learn more: goo.gle/46JPiFZ
-
-
We distinguished complex whale vocalizations using a model trained on birds. We leveraged Google DeepMind's Perch 2.0 bioacoustics foundation model's transfer learning capabilities for classifying complex whale vocalizations to accelerate marine insights. Check out the blog: goo.gle/45XKKeT
-
-
What happens when we treat AI accessibility as a foundational capability? We believe technology should be as unique as the person using it. By designing tools that dynamically adapt to a user’s explicit preferences, we create more resilient, responsive technology that works for everyone. At the A3 Business Forum recently, we introduced the Natively Adaptive Interfaces (NAI) framework — a research direction designed to create more accessible applications through multimodal AI tools. NAI bakes adaptability directly into a product’s design from the beginning. Our presentation focused on the importance of building AI systems that maximize human agency. We were thrilled to share a new film produced by BBC StoryWorks Productions and RIT/NTID, documenting the journey of developing Grammar Lab alongside the Deaf community — a stellar example of our co-design philosophy, "Nothing About Us, Without Us," in action. The human impact of this work is best seen through the story of Erin Finton, an educator at RIT/NTID. By leveraging Gemini’s multimodal capabilities and support from Google.org, Erin is empowering her students to learn in ways that truly work for them. Watch the full story: https://lnkd.in/gYd42v_v Learn more on the blog: https://goo.gle/4kptDc4
-