Institution profile

DeepMind

Industry researcheurope · gb
Official website
Research library1,250linked papers
Opportunities0open roles
Selected work

Representative Papers

The Elements of Differentiable Programming

Mar 21, 2024arXiv.org

This paper addresses the challenge of end-to-end differentiability in complex programs featuring nontrivial control flow and data structures. To this end, it introduces a probabilistic programming paradigm for differentiation, unifying optimization and probabilistic inference within a differentiable programming framework. Methodologically, it transcends conventional automatic differentiation (AD) by establishing, for the first time, a theoretical link between differentiability of control flow/data structures and uncertainty modeling—integrating AD, graphical models, convex optimization, and Bayesian inference into a cohesive differentiable program modeling framework. Key contributions include: (1) revealing that differentiable programming is fundamentally probabilistic programming—not merely gradient computation; (2) proposing the “program-as-model” design principle; and (3) establishing the first comprehensive knowledge system spanning theory, design, and applications, enabling the development of differentiable software infrastructure for large language models and foundation models.

40 citations5 influentialRead paper

AlphaEvolve: A coding agent for scientific and algorithmic discovery

Jun 16, 2025

This work addresses the challenge of enabling large language models (LLMs) to autonomously solve open-ended scientific problems and optimize critical computational infrastructure. Methodologically, it introduces AlphaEvolve—a self-coding agent framework grounded in evolutionary principles—featuring multi-LLM collaborative evolution, code-level direct editing, closed-loop feedback from multiple evaluators, program semantic equivalence verification, and formal correctness proofs. Key contributions include: (1) the first breakthrough in 56 years on the scalar multiplication lower bound for 4×4 complex matrix multiplication—reducing it from Strassen’s long-standing bound of 49 to 48 operations—and generating a novel algorithm with machine-verifiable correctness; and (2) practical deployments across Google’s large-scale computing stack, achieving state-of-the-art improvements in datacenter job scheduling, hardware accelerator circuit simplification, and base LLM training acceleration. The framework discovers multiple mathematically and algorithmically superior solutions, substantially advancing the frontier of automated scientific discovery.

11 citationsRead paper

Gaga: Group Any Gaussians via 3D-aware Memory Bank

Apr 11, 2024arXiv.org

This work addresses the challenge in open-world 3D scene reconstruction and segmentation where existing methods rely on continuous-view assumptions and struggle with inconsistency among zero-shot, class-agnostic 2D masks. We propose the first framework based on a 3D-aware memory bank, leveraging 3D Gaussian splatting representations. Through a cross-view spatial association mechanism, it performs instance-level alignment and consistency modeling of heterogeneous 2D segmentation outputs under sparse and arbitrary camera poses—fully eliminating the continuous-view constraint. Crucially, class-agnostic 2D masks are dynamically bound to 3D memory units, enabling robust 3D instance memory formation and cross-view propagation. On open-world 3D segmentation benchmarks, our method significantly outperforms state-of-the-art approaches, markedly improving mask consistency and scene understanding accuracy. This establishes a novel paradigm for real-scene 3D editing and semantic interpretation.

10 citations1 influentialRead paper

Bridging Language and Action: A Survey of Language-Conditioned Robot Manipulation

Dec 17, 2023

This work addresses the semantic gap between natural language instructions and robotic physical actions to enhance the naturalness and reliability of human-robot collaboration. We propose the first four-dimensional taxonomy for language-conditioned robotic manipulation—comprising reward shaping, policy learning, neurosymbolic AI, and foundation model–driven approaches—and systematically analyze their fundamental limitations in generalization and safety. Integrating large language models (LLMs), vision-language models (VLMs), neurosymbolic reasoning, and multimodal semantic parsing, we develop a unified analytical framework spanning semantic extraction, environmental assessment, and auxiliary task design. Our analysis rigorously characterizes the performance boundaries of each paradigm for the first time, establishing theoretical foundations and concrete technical pathways toward safe, generalizable, and interpretable language-driven robotic systems.

10 citationsRead paper

Agentic Reasoning for Large Language Models

Jan 18, 2026

This work addresses the limited capacity of large language models (LLMs) to plan, act, and learn through sustained interaction in open, dynamic environments. To overcome this, the authors propose a three-tiered reasoning framework that treats LLMs as autonomous agents, unifying single-agent foundational reasoning, self-evolution, and multi-agent collaboration within a coherent paradigm. The framework orchestrates structured interactions, incorporates memory mechanisms, enables tool use, and integrates both reinforcement learning and supervised fine-tuning. It explicitly distinguishes between in-context reasoning and post-training optimization pathways, systematically coupling cognition with action. Empirical validation across diverse domains—including scientific discovery, robotics, healthcare, autonomous research, and mathematics—demonstrates its effectiveness and highlights promising future directions such as personalized interaction, long-horizon engagement, world modeling, and scalable multi-agent training.

7 citations1 influentialRead paper
Recent publications

Latest Papers