Microsoft Research
Core Facts
Lean is based on dependent type theory and expresses proofs and programs in a unified manner.
LoRA (Low-Rank Adaptation) was proposed by Microsoft in 2021, with the core idea of freezing the weights of the pretrained model and only training injected low-rank decomposition matrices.
Lean is an interactive theorem prover developed by Microsoft Research, with the latest version currently being Lean 4.
Timeline (last 90 days)
Microsoft Research proposed SEMA-Auto, a multi-turn jailbreak attack agent with autonomous judgment capability that can decide on its own when to terminate.
YOCO (You Only Cache Once) is a decoder architecture design proposed by Microsoft Research in 2024, in which the first half of the model's layers generate a global KV cache that the second half of the layers directly reuse, thereby greatly compressing the total KV Cache and reducing Pre-fill computation.
Microsoft Research open-sourced Agensh, a decentralized multi-agent framework.
Dr. Chris White currently leads the newly established Catalyst Lab at Microsoft Research.
Microsoft Research proposed WildAWM (Agent World Model in the Wild), a synthetic pipeline that uses open-source models to automatically generate controllable computer environments and tasks for training computer agents.
In the late 1990s, Gurevich formed a team at Microsoft Research and developed Spec Explorer—a high-level executable specification tool based on abstract state machines.
Splitwise (ISCA 2024, Microsoft Research) adopts a PD disaggregation scheme with different machines and the same model, achieving 1.4x throughput and a 20% cost reduction, and up to 2.35x under the same power and cost budget.
Lean was initiated by Microsoft Research and has since developed the vast mathematics library mathlib; Coq is widely used for program correctness proofs.
Microsoft Research has released the StudentSync system, which uses public de-identified learning records to train an AI simulator for each type of student, for evaluating AI tutoring products.
In 1998, Microsoft Research's LookOut prototype attempted to determine whether it needed to check the calendar when it read about a meeting being arranged in an email.
40 more timeline events
All Facts (20)
Lean is based on dependent type theory and expresses proofs and programs in a unified manner.
90%VerifiedLoRA (Low-Rank Adaptation) was proposed by Microsoft in 2021, with the core idea of freezing the weights of the pretrained model and only training injected low-rank decomposition matrices.
90%VerifiedLean is an interactive theorem prover developed by Microsoft Research, with the latest version currently being Lean 4.
90%VerifiedLean was created in 2013 by Leonardo de Moura of Microsoft Research and has since evolved to Lean 4.
85%VerifiedLean is an interactive theorem prover developed by Leonardo de Moura at Microsoft Research, based on dependent type theory.
80%VerifiedLoRA (Low-Rank Adaptation) is a parameter-efficient fine-tuning method proposed by Microsoft Research in 2022.
80%VerifiedLoRA (Low-Rank Adaptation) was originally proposed by Microsoft Research for efficient adaptation of large language models.
80%VerifiedAutoGen was launched by Microsoft Research and focuses on multi-Agent collaboration scenarios.
80%VerifiedWhen a single code review exceeds 400 lines of code, the reviewer's defect detection rate drops sharply.
80%VerifiedResNet (Residual Network) was proposed by Microsoft Research's He Kaiming team in 2015 and solved the vanishing gradient problem in deep network training by introducing residual connections.
80%VerifiedAutoGen is a multi-Agent collaboration framework produced by Microsoft Research that enables flexible collaboration between Agents through programmable conversation patterns.
80%VerifiedZ3 is a Satisfiability Modulo Theories (SMT) solver developed by Microsoft Research.
75%VerifiedResidual connections originated from ResNet proposed by Microsoft in 2015, and alleviate the vanishing gradient problem in deep networks by adding the input directly to the output, skipping intermediate layers.
75%VerifiedData from Microsoft Research shows that the average human attention span has dropped from 12 seconds in 2000 to approximately 8 seconds in 2015.
75%VerifiedGraph RAG was popularized by Microsoft Research in 2024.
70%VerifiedLean is an interactive theorem prover developed by Microsoft Research that allows mathematicians to write proofs in a formal language and have their logical correctness verified by a computer.
65%VerifiedLean is an interactive theorem prover driven by dependent type theory, developed by Leonardo de Moura's team at Microsoft Research, with Lean 4 being the currently active version.
65%VerifiedLean is an interactive theorem prover and functional programming language led in development by Leonardo de Moura of Microsoft Research.
65%VerifiedMicrosoft Research first proposed BitNet in 2023 and introduced the evolved version BitNet b1.58 in 2024.
65%VerifiedLean是由微软研究院Leonardo de Moura开发的交互式定理证明器和函数式编程语言
65%