Engineering & computing · OECD FoS 1.2 · China MoE 0812 计算机科学与技术
Computer science (algorithms & theory)
计算机科学（算法与理论）
L3Systemic transformation ↑ Rising Score 2.90 · Last reviewed 2026-10-22 · Next review 2026-12-20
BIGGEST BREAKTHROUGH 2025-05-14 AI leapField · major
AlphaEvolve: coding agent for algorithm discovery ↗
General evolutionary coding agent beating best-known results
Why this level
AI has found adopted algorithms (AlphaDev sorting merged into LLVM libc++; AlphaTensor) and improved hardness bounds (AlphaEvolve 2026); The AI Scientist appeared in Nature (Mar 2026).
S = 0.30·method + 0.35·discovery + 0.25·adoption + 0.10·education (education scores 0 when there is no evidence; if empty, the other weights are renormalised)
Sources
- nature.com/articles/s41586-023-06004-9
- research.google/blog/ai-as-a-research-partner-advancing-theoretical-computer-science-with-alphaevolve
- nature.com/articles/s41586-026-10265-5
Composite 2.5–3.49 (or ≥3.5 with any of method/discovery/adoption below 3): AI is standard infrastructure with multiple independent breakthroughs. All levels →
Research directions
Algorithm discoveryComplexity-theory proofsAutomated AI researchCompiler & systems optimisation
Papers
- Towards end-to-end automation of AI research
End-to-end automated research in Nature; output quality limited
2026-03-25 · Nature
- Reinforced Generation of Combinatorial Structures: Hardness of Approximation
AI-found structures improve hardness bounds in a mature area, fully verifiable
2025-09-22 · arXiv
- Faster sorting algorithms discovered using deep reinforcement learning
AI-found sort routines merged into LLVM libc++
2023-06-07 · Nature
-
AI leap
Discovering faster matrix multiplication algorithms with reinforcement learningFirst improvement on Strassen for 4×4 (mod 2) in 50 years
2022-10-05 · Nature
- The Six-Dimensional Tarski Problem is $Ω(\log^3 n)$-Hard
Improves lower bounds for finding Tarski fixed points, giving optimal Θ((log n)^3) in six dimensions, after recent AI-assisted proofs.
2026-10-15 · arXiv · Kristóf Zólomy
- Online Tokenization with Optimal Delay in Linear Time
An online BPE tokenisation algorithm with optimal delay in linear time, reducing latency and memory.
2026-10-13 · arXiv · Mina Dalirrooyfard, Christopher Ye
- Fast Quantized Johnson$-$Lindenstrauss Transforms
Faster structured quantised JL transforms for 1-bit KV-cache compression in LLM inference.
2026-10-12 · arXiv · Angelos Korakitis, Vasileios Nakos
- Hashrope: A Persistent Data Structure for Mechanisms of LLM Context Management
Shows four LLM context-management mechanisms reduce to one hashed persistent weight-balanced rope.
2026-10-12 · arXiv · Muntaser Syed, Ameerah Alsulami, Marius Silaghi
- Approximation Algorithms for Dual-Memory Request Scheduling in Disaggregated LLM Inference
Formalises coupled prefill/decode memory scheduling in disaggregated LLM inference with approximation guarantees.
2026-10-11 · arXiv · Xin Wang, Hong Shen, Hui Tian, Ye Tao
- Cascade: GPU-Accelerated Out-of-Core Suffix Array Construction for LLM Corpora
GPU-accelerated out-of-core suffix-array construction for LLM corpora used in deduplication, contamination tests and attribution.
2026-10-11 · arXiv · Jamshed Khan, Prashant Pandey
- From arXiv, bioRxiv, medRxiv and OpenAlex (last ~30 days); summaries written by AIMap from the abstract.
- PAR: Progress-Aligned Recurrence for Length Generalization in State Tracking 2026-10-20
- Focused Omni-Modal GUI Agents 2026-10-19
- EgoXperience: Learning World-Grounded Egocentric Representations from Human Experience 2026-10-19
- Holi-World: Building Holistic World Intelligence from Human-Ego Experience 2026-10-19
- MemT-1.0: Towards Memory-Native Evolution 2026-10-19
- OpenVLHarness: Real World Visual Agentic Tool-Use with Multimodal Memory 2026-10-19
- All related papers in the archive →
Milestones 8· 2 leaps
Big glowing nodes are leaps: AI leap cool · Field leap warm · double ring = both
-
Paper
Quanta: as AI closed in on a Unique Games proof, researchers raced to beat itAI nearly completes a major complexity proof
Quanta Magazine -
Discovery
KLS conjecture resolved with AI-agent assistance: dimension-free guarantees for log-concave sampling and random-walk mixingAI: Microsoft says human-directed AI agents took part in the proof search (vendor claim). Field: KLS governs mixing of high-dimensional random walks, central to volume computation and log-concave sampling complexity; backed by an independent proof (arXiv 2610.05474) but not yet refereed
Microsoft ResearchPreprintVendor claim -
Paper
The AI Scientist end-to-end automated research published in NatureEnd-to-end automated research in Nature; output quality limited
Nature -
Paper
Google Research: advancing theoretical CS with AlphaEvolveOfficial write-up (date confirmed 2025-09-30)
Google Research -
Paper
AlphaEvolve gadgets improve MAX-4-CUT (0.9883→0.987) and metric TSP (117/116→111/110) inapproximabilityAI-found structures improve hardness bounds in a mature area, fully verifiable
arXivPreprint -
SystemAI leap
AlphaEvolve: coding agent for algorithm discoveryGeneral evolutionary coding agent beating best-known results
Google DeepMind -
Paper
AlphaDev finds faster sorting, merged into LLVM (Nature)AI-found sort routines merged into LLVM libc++
Nature -
PaperAI leap
AlphaTensor discovers faster matrix multiplication (Nature)First improvement on Strassen for 4×4 (mod 2) in 50 years
Nature
No milestones match this filter.
More papers & repos 6
| Type | Title | Date |
|---|---|---|
| Paper | Towards end-to-end automation of AI research | 2026-03-25 |
| Paper | Reinforced Generation of Combinatorial Structures: Hardness of Approximation Preprint | 2025-09-22 |
| Paper | Faster sorting algorithms discovered using deep reinforcement learning | 2023-06-07 |
| Paper | Discovering faster matrix multiplication algorithms with reinforcement learning | 2022-10-05 |
| Repo | SakanaAI/AI-Scientist-v2 To be verified | — |
| Repo | google-deepmind/alphaevolve_results To be verified | — |
Latest news 5
- TRACE: Governing AI-Assisted Research with Principles from Computer SciencearXiv q-finenPaperPreprint
- Discuss More, Comment Less: Boosting Automated Research via Multi-Perspective Collaborative ReviewarXiv cs.CLenPaperPreprint
- Open No MoreComputational Complexityen