Blog
데이터 과학, 인공지능, 딥러닝에 관한 이야기
163개의 포스트
인간 행동의 복잡성을 이해하기 위한 새로운 접근법
현대 사회에서 인간 행동의 복잡성은 데이터 과학과 인공지능(AI) 연구의 중요한 주제가 되고 있습니다. 특히, 이러한 연구는 우리의 일상 생활과 밀접하게 연결되어 있으며, 사회적 상호작용부터 의사 결정에 이르기까지 다양한 분야에 영향을 미칩니다. 최근 Nature와 같은 저명한 학술지에 게재되는 연구들은 이 주제에 대한 새로운 통찰을 제공합니다. 본 블로그...
![[논문 리뷰] Reasoning to Learn from Latent Thoughts](/assets/images/blog/20260223-paper-2503-18866-reasoning-to-learn-from-latent.jpg)
[논문 리뷰] Reasoning to Learn from Latent Thoughts
Compute scaling for language model (LM) pretraining has outpaced the growth of human-written texts, leading to concerns that data will become the bottleneck to LM scaling. To continue scaling pretrain...
![[논문 리뷰] Reinforcement Learning via Self-Distillation](/assets/images/blog/20260222-paper-2601-20802-reinforcement-learning-via-sel.jpg)
[논문 리뷰] Reinforcement Learning via Self-Distillation
Large language models are increasingly post-trained with reinforcement learning in verifiable domains such as code and math. Yet, current methods for reinforcement learning with verifiable rewards (RL...
![[논문 리뷰] Unified Latents (UL): How to train your latents](/assets/images/blog/20260221-paper-2602-17270-unified-latents-ul-how-to-trai.jpg)
[논문 리뷰] Unified Latents (UL): How to train your latents
We present Unified Latents (UL), a framework for learning latent representations that are jointly regularized by a diffusion prior and decoded by a diffusion model. By linking the encoder's output noi...
![[논문 리뷰] One-step Language Modeling via Continuous Denoising](/assets/images/blog/20260221-paper-2602-16813-one-step-language-modeling-via.jpg)
[논문 리뷰] One-step Language Modeling via Continuous Denoising
Language models based on discrete diffusion have attracted widespread interest for their potential to provide faster generation than autoregressive models. In practice, however, they exhibit a sharp d...
![[논문 리뷰] Towards a Science of AI Agent Reliability](/assets/images/blog/20260221-paper-2602-16666-towards-a-science-of-ai-agent-.jpg)
[논문 리뷰] Towards a Science of AI Agent Reliability
AI agents are increasingly deployed to execute important tasks. While rising accuracy scores on standard benchmarks suggest rapid progress, many agents still continue to fail in practice. This discrep...
![[논문 리뷰] Learning to Learn from Language Feedback with Social Meta-Learning](/assets/images/blog/20260221-paper-2602-16488-learning-to-learn-from-languag.jpg)
[논문 리뷰] Learning to Learn from Language Feedback with Social Meta-Learning
Large language models (LLMs) often struggle to learn from corrective feedback within a conversational context. They are rarely proactive in soliciting this feedback, even when faced with ambiguity, wh...
![[논문 리뷰] MemoryArena: Benchmarking Agent Memory in Interdependent Multi-Session Agentic Tasks](/assets/images/blog/20260221-paper-2602-16313-memoryarena-benchmarking-agent.jpg)
[논문 리뷰] MemoryArena: Benchmarking Agent Memory in Interdependent Multi-Session Agentic Tasks
Existing evaluations of agents with memory typically assess memorization and action in isolation. One class of benchmarks evaluates memorization by testing recall of past conversations or text but fai...
![[논문 리뷰] Long-Tail Knowledge in Large Language Models: Taxonomy, Mechanisms, Interventions and Implications](/assets/images/blog/20260221-paper-2602-16201-long-tail-knowledge-in-large-l.jpg)
[논문 리뷰] Long-Tail Knowledge in Large Language Models: Taxonomy, Mechanisms, Interventions and Implications
Large language models (LLMs) are trained on web-scale corpora that exhibit steep power-law distributions, in which the distribution of knowledge is highly long-tailed, with most appearing infrequently...
![[논문 리뷰] Conjugate Learning Theory: Uncovering the Mechanisms of Trainability and Generalization in Deep Neural Networks](/assets/images/blog/20260221-paper-2602-16177-conjugate-learning-theory-unco.jpg)
[논문 리뷰] Conjugate Learning Theory: Uncovering the Mechanisms of Trainability and Generalization in Deep Neural Networks
In this work, we propose a notion of practical learnability grounded in finite sample settings, and develop a conjugate learning theoretical framework based on convex conjugate duality to characterize...
![[논문 리뷰] Learning Personalized Agents from Human Feedback](/assets/images/blog/20260221-paper-2602-16173-learning-personalized-agents-f.jpg)
[논문 리뷰] Learning Personalized Agents from Human Feedback
Modern AI agents are powerful but often fail to align with the idiosyncratic, evolving preferences of individual users. Prior approaches typically rely on static datasets, either training implicit pre...
![[논문 리뷰] Scaling Beyond Masked Diffusion Language Models](/assets/images/blog/20260221-paper-2602-15014-scaling-beyond-masked-diffusio.jpg)
[논문 리뷰] Scaling Beyond Masked Diffusion Language Models
Diffusion language models are a promising alternative to autoregressive models due to their potential for faster generation. Among discrete diffusion approaches, Masked diffusion currently dominates, ...
![[논문 리뷰] When Models Manipulate Manifolds: The Geometry of a Counting Task](/assets/images/blog/20260221-paper-2601-04480-when-models-manipulate-manifol.jpg)
[논문 리뷰] When Models Manipulate Manifolds: The Geometry of a Counting Task
Language models can perceive visual properties of text despite receiving only sequences of tokens-we mechanistically investigate how Claude 3.5 Haiku accomplishes one such task: linebreaking in fixed-...
![[논문 리뷰] Knowledge Entropy Decay during Language Model Pretraining Hinders New Knowledge Acquisition](/assets/images/blog/20260221-paper-2410-01380-knowledge-entropy-decay-during.jpg)
[논문 리뷰] Knowledge Entropy Decay during Language Model Pretraining Hinders New Knowledge Acquisition
In this work, we investigate how a model's tendency to broadly integrate its parametric knowledge evolves throughout pretraining, and how this behavior affects overall performance, particularly in ter...
![[논문 리뷰] GLM-5: from Vibe Coding to Agentic Engineering](/assets/images/blog/20260220-paper-2602-15763-glm-5-from-vibe-coding-to-agen.jpg)
[논문 리뷰] GLM-5: from Vibe Coding to Agentic Engineering
We present GLM-5, a next-generation foundation model designed to transition the paradigm of vibe coding to agentic engineering. Building upon the agentic, reasoning, and coding (ARC) capabilities of i...
![[논문 리뷰] PaperBanana: Automating Academic Illustration for AI Scientists](/assets/images/blog/20260219-paper-2601-23265-paperbanana-automating-academi.jpg)
[논문 리뷰] PaperBanana: Automating Academic Illustration for AI Scientists
Despite rapid advances in autonomous AI scientists powered by language models, generating publication-ready illustrations remains a labor-intensive bottleneck in the research workflow. To lift this bu...
![[논문 리뷰] On-Policy Context Distillation for Language Models](/assets/images/blog/20260218-paper-2602-12275-on-policy-context-distillation.jpg)
[논문 리뷰] On-Policy Context Distillation for Language Models
Context distillation enables language models to internalize in-context knowledge into their parameters. In our work, we propose On-Policy Context Distillation (OPCD), a framework that bridges on-polic...
![[논문 리뷰] Evolutionary Router Feature Generation for Zero-Shot Graph Anomaly Detection with Mixture-of-Experts](/assets/images/blog/20260218-paper-2602-11622-evolutionary-router-feature-ge.jpg)
[논문 리뷰] Evolutionary Router Feature Generation for Zero-Shot Graph Anomaly Detection with Mixture-of-Experts
Zero-shot graph anomaly detection (GAD) has attracted increasing attention recent years, yet the heterogeneity of graph structures, features, and anomaly patterns across graphs make existing single GN...
![[논문 리뷰] The Implicit Bias of Steepest Descent with Mini-batch Stochastic Gradient](/assets/images/blog/20260218-paper-2602-11557-the-implicit-bias-of-steepest-.jpg)
[논문 리뷰] The Implicit Bias of Steepest Descent with Mini-batch Stochastic Gradient
A variety of widely used optimization methods like SignSGD and Muon can be interpreted as instances of steepest descent under different norm-induced geometries. In this work, we study the implicit bia...
![[논문 리뷰] RiemannGL: Riemannian Geometry Changes Graph Deep Learning](/assets/images/blog/20260218-paper-2602-10982-riemanngl-riemannian-geometry-.jpg)
[논문 리뷰] RiemannGL: Riemannian Geometry Changes Graph Deep Learning
Graphs are ubiquitous, and learning on graphs has become a cornerstone in artificial intelligence and data mining communities. Unlike pixel grids in images or sequential structures in language, graphs...
![[논문 리뷰] Towards Autonomous Mathematics Research](/assets/images/blog/20260218-paper-2602-10177-towards-autonomous-mathematics.jpg)
[논문 리뷰] Towards Autonomous Mathematics Research
Recent advances in foundational models have yielded reasoning systems capable of achieving a gold-medal standard at the International Mathematical Olympiad. The transition from competition-level probl...
![[논문 리뷰] Theory of Space: Can Foundation Models Construct Spatial Beliefs through Active Exploration?](/assets/images/blog/20260218-paper-2602-07055-theory-of-space-can-foundation.jpg)
[논문 리뷰] Theory of Space: Can Foundation Models Construct Spatial Beliefs through Active Exploration?
Spatial embodied intelligence requires agents to act to acquire information under partial observability. While multimodal foundation models excel at passive perception, their capacity for active, self...
![[논문 리뷰] HyperMLP: An Integrated Perspective for Sequence Modeling](/assets/images/blog/20260217-paper-2602-12601-hypermlp-an-integrated-perspec.jpg)
[논문 리뷰] HyperMLP: An Integrated Perspective for Sequence Modeling
Self-attention is often viewed as probabilistic query-key lookup, motivating designs that preserve normalized attention scores and fixed positional semantics. We advocate a simpler and more unified pe...
![[논문 리뷰] Think like a Scientist: Physics-guided LLM Agent for Equation Discovery](/assets/images/blog/20260217-paper-2602-12259-think-like-a-scientist-physics.jpg)
[논문 리뷰] Think like a Scientist: Physics-guided LLM Agent for Equation Discovery
Explaining observed phenomena through symbolic, interpretable formulas is a fundamental goal of science. Recently, large language models (LLMs) have emerged as promising tools for symbolic equation di...
![[논문 리뷰] Latent Forcing: Reordering the Diffusion Trajectory for Pixel-Space Image Generation](/assets/images/blog/20260217-paper-2602-11401-latent-forcing-reordering-the-.jpg)
[논문 리뷰] Latent Forcing: Reordering the Diffusion Trajectory for Pixel-Space Image Generation
Latent diffusion models excel at generating high-quality images but lose the benefits of end-to-end modeling. They discard information during image encoding, require a separately trained decoder, and ...
![[논문 리뷰] Causal-JEPA: Learning World Models through Object-Level Latent Interventions](/assets/images/blog/20260217-paper-2602-11389-causal-jepa-learning-world-mod.jpg)
[논문 리뷰] Causal-JEPA: Learning World Models through Object-Level Latent Interventions
World models require robust relational understanding to support prediction, reasoning, and control. While object-centric representations provide a useful abstraction, they are not sufficient to captur...
![[논문 리뷰] TabICLv2: A better, faster, scalable, and open tabular foundation model](/assets/images/blog/20260217-paper-2602-11139-tabiclv2-a-better-faster-scala.jpg)
[논문 리뷰] TabICLv2: A better, faster, scalable, and open tabular foundation model
Tabular foundation models, such as TabPFNv2 and TabICL, have recently dethroned gradient-boosted trees at the top of predictive benchmarks, demonstrating the value of in-context learning for tabular d...
![[논문 리뷰] VLA-JEPA: Enhancing Vision-Language-Action Model with Latent World Model](/assets/images/blog/20260217-paper-2602-10098-vla-jepa-enhancing-vision-lang.jpg)
[논문 리뷰] VLA-JEPA: Enhancing Vision-Language-Action Model with Latent World Model
Pretraining Vision-Language-Action (VLA) policies on internet-scale video is appealing, yet current latent-action objectives often learn the wrong thing: they remain anchored to pixel variation rather...
AI와 데이터 과학의 미래: 최전선의 기술과 혁신
인공지능(AI)과 데이터 과학은 현대 기술 혁신을 주도하는 두 개의 거대한 축입니다. 이들은 우리의 일상생활을 넘어, 의료, 금융, 제조 등 산업 전반에 걸쳐 근본적인 변화를 일으키고 있습니다. AI는 인간의 지능을 모방하여 자율주행차를 운행하고, 질병을 진단하며, 개인에게 맞춤형 콘텐츠를 추천합니다. 데이터 과학은 이러한 AI 모델이 최상의 성능을 내도록...
![[논문 리뷰] Agent World Model: Infinity Synthetic Environments for Agentic Reinforcement Learning](/assets/images/blog/20260213-paper-2602-10090-agent-world-model-infinity-syn.jpg)
[논문 리뷰] Agent World Model: Infinity Synthetic Environments for Agentic Reinforcement Learning
Recent advances in large language model (LLM) have empowered autonomous agents to perform complex tasks that require multi-turn interactions with tools and environments. However, scaling such agent tr...
![[논문 리뷰] Deriving Neural Scaling Laws from the statistics of natural language](/assets/images/blog/20260213-paper-2602-07488-deriving-neural-scaling-laws-f.jpg)
[논문 리뷰] Deriving Neural Scaling Laws from the statistics of natural language
Despite the fact that experimental neural scaling laws have substantially guided empirical progress in large-scale machine learning, no existing theory can quantitatively predict the exponents of thes...
![[논문 리뷰] Research on World Models Is Not Merely Injecting World Knowledge into Specific Tasks](/assets/images/blog/20260213-paper-2602-01630-research-on-world-models-is-no.jpg)
[논문 리뷰] Research on World Models Is Not Merely Injecting World Knowledge into Specific Tasks
World models have emerged as a critical frontier in AI research, aiming to enhance large models by infusing them with physical dynamics and world knowledge. The core objective is to enable agents to u...
![[논문 리뷰] Recursive Language Models](/assets/images/blog/20260213-paper-2512-24601-recursive-language-models.jpg)
[논문 리뷰] Recursive Language Models
We study allowing large language models (LLMs) to process arbitrarily long prompts through the lens of inference-time scaling. We propose Recursive Language Models (RLMs), a general inference paradigm...
자가 지도 학습의 발전: 데이터 효율적인 학습을 향한 여정
현대의 인공지능(AI) 기술은 대부분 대량의 레이블이 있는 데이터에 의존하여 모델을 학습시킵니다. 그러나 현실 세계의 데이터 대부분은 레이블이 없으며, 수동으로 레이블을 만드는 작업은 막대한 비용과 시간이 소요됩니다. 이러한 문제를 해결하기 위한 강력한 접근 방식인 자가 지도 학습(Self-Supervised Learning, SSL)은 최근 AI 연구의 ...
![[논문 리뷰] Reinforced Attention Learning](/assets/images/blog/20260211-paper-2602-04884-reinforced-attention-learning.jpg)
[논문 리뷰] Reinforced Attention Learning
Post-training with Reinforcement Learning (RL) has substantially improved reasoning in Large Language Models (LLMs) via test-time scaling. However, extending this paradigm to Multimodal LLMs (MLLMs) t...
인간 행동 모델링의 새로운 패러다임: 사회적 상호작용에 대한 심층 학습 접근 방법
인간 행동 모델링은 인공지능(AI)과 데이터 과학 분야에서 가장 흥미롭고 도전적인 연구 주제 중 하나입니다. 특히 인간의 사회적 상호작용을 이해하고 예측하는 능력은 대화형 AI, 로보틱스, 정신 건강 케어 등 다양한 응용 분야에서 혁신을 가져올 큰 잠재력을 가지고 있습니다. 최근 Nature에 발표된 연구(https://www.nature.com/artic...
![[논문 리뷰] PluRel: Synthetic Data unlocks Scaling Laws for Relational Foundation Models](/assets/images/blog/20260210-paper-2602-04029-plurel-synthetic-data-unlocks-.jpg)
[논문 리뷰] PluRel: Synthetic Data unlocks Scaling Laws for Relational Foundation Models
Relational Foundation Models (RFMs) facilitate data-driven decision-making by learning from complex multi-table databases. However, the diverse relational databases needed to train such models are rar...
![[논문 리뷰] Learning to Reason in 13 Parameters](/assets/images/blog/20260209-paper-2602-04118-learning-to-reason-in-13-param.jpg)
[논문 리뷰] Learning to Reason in 13 Parameters
Recent research has shown that language models can learn to extit{reason}, often via reinforcement learning. Some work even trains low-rank parameterizations for reasoning, but conventional LoRA can...
![[논문 리뷰] Titans: Learning to Memorize at Test Time](/assets/images/blog/20260205-paper-2501-00663-titans-learning-to-memorize-at.jpg)
[논문 리뷰] Titans: Learning to Memorize at Test Time
Over more than a decade there has been an extensive research effort on how to effectively utilize recurrent models and attention. While recurrent models aim to compress the data into a fixed-size memo...
![[논문 리뷰] Knowledge Graphs are Implicit Reward Models: Path-Derived Signals Enable Compositional Reasoning](/assets/images/blog/20260203-paper-2601-15160-knowledge-graphs-are-implicit-.jpg)
[논문 리뷰] Knowledge Graphs are Implicit Reward Models: Path-Derived Signals Enable Compositional Reasoning
Large language models have achieved near-expert performance in structured reasoning domains like mathematics and programming, yet their ability to perform compositional multi-hop reasoning in speciali...
![[논문 리뷰] Dynamic Sparse Learning: A Novel Paradigm for Efficient Recommendation](/assets/images/blog/20260202-paper-2402-02855-dynamic-sparse-learning-a-nove.jpg)
[논문 리뷰] Dynamic Sparse Learning: A Novel Paradigm for Efficient Recommendation
In the realm of deep learning-based recommendation systems, the increasing computational demands, driven by the growing number of users and items, pose a significant challenge to practical deployment....
![[논문 리뷰] Teaching Models to Teach Themselves: Reasoning at the Edge of Learnability](/assets/images/blog/20260201-paper-2601-18778-teaching-models-to-teach-thems.jpg)
[논문 리뷰] Teaching Models to Teach Themselves: Reasoning at the Edge of Learnability
Can a model learn to escape its own learning plateau? Reinforcement learning methods for finetuning large reasoning models stall on datasets with low initial success rates, and thus little training si...
![[논문 리뷰] On First-Order Meta-Learning Algorithms](/assets/images/blog/20260201-paper-1803-02999-on-first-order-meta-learning-a.jpg)
[논문 리뷰] On First-Order Meta-Learning Algorithms
This paper considers meta-learning problems, where there is a distribution of tasks, and we would like to obtain an agent that performs well (i.e., learns quickly) when presented with a previously uns...
![[논문 리뷰] From Seed AI to Technological Singularity via Recursively Self-Improving Software](/assets/images/blog/20260201-paper-1502-06512-from-seed-ai-to-technological-.jpg)
[논문 리뷰] From Seed AI to Technological Singularity via Recursively Self-Improving Software
Software capable of improving itself has been a dream of computer scientists since the inception of the field. In this work we provide definitions for Recursively Self-Improving software, survey diffe...
![[논문 리뷰] Resonant Sparse Geometry Networks](/assets/images/blog/20260131-paper-2601-18064-resonant-sparse-geometry-netwo.jpg)
[논문 리뷰] Resonant Sparse Geometry Networks
We introduce Resonant Sparse Geometry Networks (RSGN), a brain-inspired architecture with self-organizing sparse hierarchical input-dependent connectivity. Unlike Transformer architectures that employ...
![[논문 리뷰] Alignment Pretraining: AI Discourse Causes Self-Fulfilling (Mis)alignment](/assets/images/blog/20260130-paper-2601-10160-alignment-pretraining-ai-disco.jpg)
[논문 리뷰] Alignment Pretraining: AI Discourse Causes Self-Fulfilling (Mis)alignment
Pretraining corpora contain extensive discourse about AI systems, yet the causal influence of this discourse on downstream alignment remains poorly understood. If prevailing descriptions of AI behavio...
![[논문 리뷰] LLM-in-Sandbox Elicits General Agentic Intelligence](/assets/images/blog/20260129-paper-2601-16206-llm-in-sandbox-elicits-general.jpg)
[논문 리뷰] LLM-in-Sandbox Elicits General Agentic Intelligence
We introduce LLM-in-Sandbox, enabling LLMs to explore within a code sandbox (i.e., a virtual computer), to elicit general intelligence in non-code domains. We first demonstrate that strong LLMs, witho...
![[논문 리뷰] Large Language Model Agent: A Survey on Methodology, Applications and Challenges](/assets/images/blog/20260128-paper-2503-21460-large-language-model-agent-a-s.jpg)
[논문 리뷰] Large Language Model Agent: A Survey on Methodology, Applications and Challenges
The era of intelligent agents is upon us, driven by revolutionary advancements in large language models. Large Language Model (LLM) agents, with goal-driven behaviors and dynamic adaptation capabiliti...
![[논문 리뷰] Aligning Large Language Models to a Domain-specific Graph Database for NL2GQL](/assets/images/blog/20260128-paper-2402-16567-aligning-large-language-models.jpg)
[논문 리뷰] Aligning Large Language Models to a Domain-specific Graph Database for NL2GQL
Graph Databases (Graph DB) find extensive application across diverse domains such as finance, social networks, and medicine. Yet, the translation of Natural Language (NL) into the Graph Query Language...
TabPFN: 데이터 과학의 새로운 패러다임
오늘날 데이터 과학과 인공지능(AI)은 다양한 산업과 학문 분야에서 혁신을 주도하고 있습니다. 이러한 변화 속에서 데이터 모델링과 예측은 매우 중요한 역할을 하고 있으며, 특히 테이블 형식의 데이터를 다루는 기술은 많은 주목을 받고 있습니다. 이번 글에서는 TabPFN이라는 혁신적인 접근 방식을 소개하고, 이 기술이 데이터 과학의 미래에 어떻게 기여할 수 ...
![[논문 리뷰] Agentic Reasoning for Large Language Models](/assets/images/blog/20260125-paper-2601-12538-agentic-reasoning.jpg)
[논문 리뷰] Agentic Reasoning for Large Language Models
Reasoning is a fundamental cognitive process underlying inference, problem-solving, and decision-making. While large language models (LLMs) demonstrate strong reasoning capabilities in closed-world se...
데이터 과학 프로젝트의 성능을 높이는 방법: tuneTable
데이터 과학과 인공지능(AI) 프로젝트를 진행하다 보면 모델의 성능을 최적화하는 것이 가장 큰 도전 중 하나입니다. 어떤 알고리즘을 사용하든, 적절한 하이퍼파라미터(hyperparameter)를 찾는 과정은 모델의 성패를 좌우할 수 있습니다. 이 글에서는 그러한 최적화 문제를 해결하는 데 도움을 줄 수 있는 도구인 tuneTable에 대해 소개하고자 합니다...
![[논문 리뷰] Learning to Discover at Test Time](/assets/images/blog/20260123-paper-2601-16175-learning-to-discover-at-test-t.jpg)
[논문 리뷰] Learning to Discover at Test Time
How can we use AI to discover a new state of the art for a scientific problem? Prior work in test-time scaling, such as AlphaEvolve, performs search by prompting a frozen LLM. We perform reinforcement...
![[논문 리뷰] STEM: Scaling Transformers with Embedding Modules](/assets/images/blog/20260123-paper-2601-10639-stem-scaling-transformers-with.jpg)
[논문 리뷰] STEM: Scaling Transformers with Embedding Modules
Fine-grained sparsity promises higher parametric capacity without proportional per-token compute, but often suffers from training instability, load balancing, and communication overhead. We introduce ...
![[논문 리뷰] Aligning Agentic World Models via Knowledgeable Experience Learning](/assets/images/blog/20260122-paper-2601-13247-aligning-agentic-world-models-.jpg)
[논문 리뷰] Aligning Agentic World Models via Knowledgeable Experience Learning
Current Large Language Models (LLMs) exhibit a critical modal disconnect: they possess vast semantic knowledge but lack the procedural grounding to respect the immutable laws of the physical world. Co...
![[논문 리뷰] The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models](/assets/images/blog/20260122-paper-2601-10387-the-assistant-axis-situating-a.jpg)
[논문 리뷰] The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models
Large language models can represent a variety of personas but typically default to a helpful Assistant identity cultivated during post-training. We investigate the structure of the space of model pers...
![[논문 리뷰] Multiplex Thinking: Reasoning via Token-wise Branch-and-Merge](/assets/images/blog/20260122-paper-2601-08808-multiplex-thinking-reasoning-v.jpg)
[논문 리뷰] Multiplex Thinking: Reasoning via Token-wise Branch-and-Merge
Large language models often solve complex reasoning tasks more effectively with Chain-of-Thought (CoT), but at the cost of long, low-bandwidth token sequences. Humans, by contrast, often reason softly...
![[논문 리뷰] MetaEmbed: Scaling Multimodal Retrieval at Test-Time with Flexible Late Interaction](/assets/images/blog/20260122-paper-2509-18095-metaembed-scaling-multimodal-r.jpg)
[논문 리뷰] MetaEmbed: Scaling Multimodal Retrieval at Test-Time with Flexible Late Interaction
Universal multimodal embedding models have achieved great success in capturing semantic relevance between queries and candidates. However, current methods either condense queries and candidates into a...
![[논문 리뷰] SAGE: Tool-Augmented LLM Task Solving Strategies in Scalable Multi-Agent Environments](/assets/images/blog/20260121-paper-2601-09750-sage-tool-augmented-llm-task-s.jpg)
[논문 리뷰] SAGE: Tool-Augmented LLM Task Solving Strategies in Scalable Multi-Agent Environments
Large language models (LLMs) have proven to work well in question-answering scenarios, but real-world applications often require access to tools for live information or actuation. For this, LLMs can b...
![[논문 리뷰] Urban Socio-Semantic Segmentation with Vision-Language Reasoning](/assets/images/blog/20260119-paper-2601-10477-urban-socio-semantic-segmentat.jpg)
[논문 리뷰] Urban Socio-Semantic Segmentation with Vision-Language Reasoning
As hubs of human activity, urban surfaces consist of a wealth of semantic entities. Segmenting these various entities from satellite imagery is crucial for a range of downstream applications. Current ...
![[논문 리뷰] KGGen: Extracting Knowledge Graphs from Plain Text with Language Models](/assets/images/blog/20260118-paper-2502-09956-kggen-extracting-knowledge-gra.jpg)
[논문 리뷰] KGGen: Extracting Knowledge Graphs from Plain Text with Language Models
Recent interest in building foundation models for KGs has highlighted a fundamental challenge: knowledge-graph data is relatively scarce. The best-known KGs are primarily human-labeled, created by pat...
![[논문 리뷰] Agentic Memory: Learning Unified Long-Term and Short-Term Memory Management for Large Language Model Agents](/assets/images/blog/20260115-paper-2601-01885-agentic-memory-learning-unifie.jpg)
[논문 리뷰] Agentic Memory: Learning Unified Long-Term and Short-Term Memory Management for Large Language Model Agents
Large language model (LLM) agents face fundamental limitations in long-horizon reasoning due to finite context windows, making effective memory management critical. Existing methods typically handle l...
![[논문 리뷰] Tracing Moral Foundations in Large Language Models](/assets/images/blog/20260113-paper-2601-05437-tracing-moral-foundations-in-l.jpg)
[논문 리뷰] Tracing Moral Foundations in Large Language Models
Large language models (LLMs) often produce human-like moral judgments, but it is unclear whether this reflects an internal conceptual structure or superficial ``moral mimicry.'' Using Moral Foundation...
![[논문 리뷰] From Entropy to Epiplexity: Rethinking Information for Computationally Bounded Intelligence](/assets/images/blog/20260113-paper-2601-03220-from-entropy-to-epiplexity-ret.jpg)
[논문 리뷰] From Entropy to Epiplexity: Rethinking Information for Computationally Bounded Intelligence
Can we learn more from data than existed in the generating process itself? Can new and useful information be constructed from merely applying deterministic transformations to existing data? Can the le...
![[논문 리뷰] Token-Level LLM Collaboration via FusionRoute](/assets/images/blog/20260112-paper-2601-05106-token-level-llm-collaboration-.jpg)
[논문 리뷰] Token-Level LLM Collaboration via FusionRoute
Large language models (LLMs) exhibit strengths across diverse domains. However, achieving strong performance across these domains with a single general-purpose model typically requires scaling to size...
![[논문 리뷰] GDPO: Group reward-Decoupled Normalization Policy Optimization for Multi-reward RL Optimization](/assets/images/blog/20260111-paper-2601-05242-gdpo-group-reward-decoupled-no.jpg)
[논문 리뷰] GDPO: Group reward-Decoupled Normalization Policy Optimization for Multi-reward RL Optimization
As language models become increasingly capable, users expect them to provide not only accurate responses but also behaviors aligned with diverse human preferences across a variety of scenarios. To ach...
![[논문 리뷰] Learning Latent Action World Models In The Wild](/assets/images/blog/20260111-paper-2601-05230-learning-latent-action-world-m.jpg)
[논문 리뷰] Learning Latent Action World Models In The Wild
Agents capable of reasoning and planning in the real world require the ability of predicting the consequences of their actions. While world models possess this capability, they most often require acti...
![[논문 리뷰] Jenius Agent: Towards Experience-Driven Accuracy Optimization in Real-World Scenarios](/assets/images/blog/20260111-paper-2601-01857-jenius-agent-towards-experienc.jpg)
[논문 리뷰] Jenius Agent: Towards Experience-Driven Accuracy Optimization in Real-World Scenarios
As agent systems powered by large language models (LLMs) advance, improving the task performance of an autonomous agent, especially in context understanding, tool usage, and response generation, has b...
![[논문 리뷰] End-to-End Test-Time Training for Long Context](/assets/images/blog/20260111-paper-2512-23675-end-to-end-test-time-training-.jpg)
[논문 리뷰] End-to-End Test-Time Training for Long Context
We formulate long-context language modeling as a problem in continual learning rather than architecture design. Under this formulation, we only use a standard architecture -- a Transformer with slidin...
![[논문 리뷰] The Missing Layer of AGI: From Pattern Alchemy to Coordination Physics](/assets/images/blog/20260111-paper-2512-05765-the-missing-layer-of-agi-from-.jpg)
[논문 리뷰] The Missing Layer of AGI: From Pattern Alchemy to Coordination Physics
Influential critiques argue that Large Language Models (LLMs) are a dead end for AGI: "mere pattern matchers" structurally incapable of reasoning or planning. We argue this conclusion misidentifies th...
![[논문 리뷰] Confucius Code Agent: Scalable Agent Scaffolding for Real-World Codebases](/assets/images/blog/20260110-paper-2512-10398-confucius-code-agent-scalable-.jpg)
[논문 리뷰] Confucius Code Agent: Scalable Agent Scaffolding for Real-World Codebases
Real-world software engineering tasks require coding agents that can operate over massive repositories, sustain long-horizon sessions, and reliably coordinate complex toolchains at test time. Existing...
![[논문 리뷰] RelayLLM: Efficient Reasoning via Collaborative Decoding](/assets/images/blog/20260109-paper-2601-05167-relayllm-efficient-reasoning-v.jpg)
[논문 리뷰] RelayLLM: Efficient Reasoning via Collaborative Decoding
Large Language Models (LLMs) for complex reasoning is often hindered by high computational costs and latency, while resource-efficient Small Language Models (SLMs) typically lack the necessary reasoni...
![[논문 리뷰] AT²PO: Agentic Turn-based Policy Optimization via Tree Search](/assets/images/blog/20260109-paper-2601-04767-at-2-po-agentic-turn-based-pol.jpg)
[논문 리뷰] AT²PO: Agentic Turn-based Policy Optimization via Tree Search
LLM agents have emerged as powerful systems for tackling multi-turn tasks by interleaving internal reasoning and external tool interactions. Agentic Reinforcement Learning has recently drawn significa...
![[논문 리뷰] 1000 Layer Networks for Self-Supervised RL: Scaling Depth Can Enable New Goal-Reaching Capabilities](/assets/images/blog/20260109-paper-2503-14858-1000-layer-networks-for-self-s.jpg)
[논문 리뷰] 1000 Layer Networks for Self-Supervised RL: Scaling Depth Can Enable New Goal-Reaching Capabilities
Scaling up self-supervised learning has driven breakthroughs in language and vision, yet comparable progress has remained elusive in reinforcement learning (RL). In this paper, we study building block...
![[논문 리뷰] Extracting books from production language models](/assets/images/blog/20260108-paper-2601-02671-extracting-books-from-producti.jpg)
[논문 리뷰] Extracting books from production language models
Many unresolved legal questions over LLMs and copyright center on memorization: whether specific training data have been encoded in the model's weights during training, and whether those memorized dat...
![[논문 리뷰] The Platonic Representation Hypothesis](/assets/images/blog/20260106-paper-2405-07987-the-platonic-representation-hy.jpg)
[논문 리뷰] The Platonic Representation Hypothesis
We argue that representations in AI models, particularly deep networks, are converging. First, we survey many examples of convergence in the literature: over time and across multiple domains, the ways...
![[논문 리뷰] Deep Delta Learning](/assets/images/blog/20260104-paper-url-pdf-deep-delta-learning.jpg)
[논문 리뷰] Deep Delta Learning
The efficacy of deep residual networks is fundamentally predicated on the identity shortcut connection. While this mechanism effectively mitigates the vanishing gradient problem, it imposes a strictly...
![[논문 리뷰] SeedFold: Scaling Biomolecular Structure Prediction](/assets/images/blog/20260104-paper-2512-24354-seedfold-scaling-biomolecular-.jpg)
[논문 리뷰] SeedFold: Scaling Biomolecular Structure Prediction
Highly accurate biomolecular structure prediction is a key component of developing biomolecular foundation models, and one of the most critical aspects of building foundation models is identifying the...
![[논문 리뷰] CogRec: A Cognitive Recommender Agent Fusing Large Language Models and Soar for Explainable Recommendation](/assets/images/blog/20260104-paper-2512-24113-cogrec-a-cognitive-recommender.jpg)
[논문 리뷰] CogRec: A Cognitive Recommender Agent Fusing Large Language Models and Soar for Explainable Recommendation
Large Language Models (LLMs) have demonstrated a remarkable capacity in understanding user preferences for recommendation systems. However, they are constrained by several critical challenges, includi...
![[논문 리뷰] Improving Multi-step RAG with Hypergraph-based Memory for Long-Context Complex Relational Modeling](/assets/images/blog/20260104-paper-2512-23959-improving-multi-step-rag-with-.jpg)
[논문 리뷰] Improving Multi-step RAG with Hypergraph-based Memory for Long-Context Complex Relational Modeling
Multi-step retrieval-augmented generation (RAG) has become a widely adopted strategy for enhancing large language models (LLMs) on tasks that demand global comprehension and intensive reasoning. Many ...
![[논문 리뷰] Training AI Co-Scientists Using Rubric Rewards](/assets/images/blog/20260104-paper-2512-23707-training-ai-co-scientists-usin.jpg)
[논문 리뷰] Training AI Co-Scientists Using Rubric Rewards
AI co-scientists are emerging as a tool to assist human researchers in achieving their research goals. A crucial feature of these AI co-scientists is the ability to generate a research plan given a se...
![[논문 리뷰] SPIRAL: Symbolic LLM Planning via Grounded and Reflective Search](/assets/images/blog/20260104-paper-2512-23167-spiral-symbolic-llm-planning-v.jpg)
[논문 리뷰] SPIRAL: Symbolic LLM Planning via Grounded and Reflective Search
Large Language Models (LLMs) often falter at complex planning tasks that require exploration and self-correction, as their linear reasoning process struggles to recover from early mistakes. While sear...
![[논문 리뷰] Attention Is Not What You Need](/assets/images/blog/20260104-paper-2512-19428-attention-is-not-what-you-need.jpg)
[논문 리뷰] Attention Is Not What You Need
We revisit a basic question in sequence modeling: is explicit self-attention actually necessary for strong performance and reasoning? We argue that standard multi-head attention is best seen as a form...
![[논문 리뷰] Scaling and context steer LLMs along the same computational path as the human brain](/assets/images/blog/20260104-paper-2512-01591-scaling-and-context-steer-llms.jpg)
[논문 리뷰] Scaling and context steer LLMs along the same computational path as the human brain
Recent studies suggest that the representations learned by large language models (LLMs) are partially aligned to those of the human brain. However, whether and why this alignment score arises from a s...
![[논문 리뷰] Zero-Overhead Introspection for Adaptive Test-Time Compute](/assets/images/blog/20260104-paper-2512-01457-zero-overhead-introspection-fo.jpg)
[논문 리뷰] Zero-Overhead Introspection for Adaptive Test-Time Compute
Large language models excel at reasoning but lack key aspects of introspection, including anticipating their own success and the computation required to achieve it. Humans use real-time introspection ...
![[논문 리뷰] A Survey on Large Language Models for Mathematical Reasoning](/assets/images/blog/20260104-paper-2506-08446-a-survey-on-large-language-mod.jpg)
[논문 리뷰] A Survey on Large Language Models for Mathematical Reasoning
Mathematical reasoning has long represented one of the most fundamental and challenging frontiers in artificial intelligence research. In recent years, large language models (LLMs) have achieved signi...
![[논문 리뷰] Aligning machine and human visual representations across abstraction levels](/assets/images/blog/20260103-paper-url-pdf-aligning-machine-and-human-vis.jpg)
[논문 리뷰] Aligning machine and human visual representations across abstraction levels
...
![[논문 리뷰] Completed Hyperparameter Transfer across Modules, Width, Depth, Batch and Duration](/assets/images/blog/20260103-paper-2512-22382-completed-hyperparameter-trans.jpg)
[논문 리뷰] Completed Hyperparameter Transfer across Modules, Width, Depth, Batch and Duration
Hyperparameter tuning can dramatically impact training stability and final performance of large-scale models. Recent works on neural network parameterisations, such as $μ$P, have enabled transfer of o...
![[논문 리뷰] Shape of Thought: When Distribution Matters More than Correctness in Reasoning Tasks](/assets/images/blog/20260103-paper-2512-22255-shape-of-thought-when-distribu.jpg)
[논문 리뷰] Shape of Thought: When Distribution Matters More than Correctness in Reasoning Tasks
We present the surprising finding that a language model's reasoning capabilities can be improved by training on synthetic datasets of chain-of-thought (CoT) traces from more capable models, even when ...
![[논문 리뷰] SemanticGen: Video Generation in Semantic Space](/assets/images/blog/20260103-paper-2512-20619-semanticgen-video-generation-i.jpg)
[논문 리뷰] SemanticGen: Video Generation in Semantic Space
State-of-the-art video generative models typically learn the distribution of video latents in the VAE space and map them to pixels using a VAE decoder. While this approach can generate high-quality vi...
![[논문 리뷰] The Prism Hypothesis: Harmonizing Semantic and Pixel Representations via Unified Autoencoding](/assets/images/blog/20260103-paper-2512-19693-the-prism-hypothesis-harmonizi.jpg)
[논문 리뷰] The Prism Hypothesis: Harmonizing Semantic and Pixel Representations via Unified Autoencoding
Deep representations across modalities are inherently intertwined. In this paper, we systematically analyze the spectral characteristics of various semantic and pixel encoders. Interestingly, our stud...
![[논문 리뷰] Epistemological Fault Lines Between Human and Artificial Intelligence](/assets/images/blog/20260103-paper-2512-19466-epistemological-fault-lines-be.jpg)
[논문 리뷰] Epistemological Fault Lines Between Human and Artificial Intelligence
Large language models (LLMs) are widely described as artificial intelligence, yet their epistemic profile diverges sharply from human cognition. Here we show that the apparent alignment between human ...
![[논문 리뷰] Learning Hierarchical Procedural Memory for LLM Agents through Bayesian Selection and Contrastive Refinement](/assets/images/blog/20260103-paper-2512-18950-learning-hierarchical-procedur.jpg)
[논문 리뷰] Learning Hierarchical Procedural Memory for LLM Agents through Bayesian Selection and Contrastive Refinement
We present MACLA, a framework that decouples reasoning from learning by maintaining a frozen large language model while performing all adaptation in an external hierarchical procedural memory. MACLA e...
![[논문 리뷰] Toward Training Superintelligent Software Agents through Self-Play SWE-RL](/assets/images/blog/20260103-paper-2512-18552-toward-training-superintellige.jpg)
[논문 리뷰] Toward Training Superintelligent Software Agents through Self-Play SWE-RL
While current software agents powered by large language models (LLMs) and agentic reinforcement learning (RL) can boost programmer productivity, their training data (e.g., GitHub issues and pull reque...
![[논문 리뷰] Sophia: A Persistent Agent Framework of Artificial Life](/assets/images/blog/20260103-paper-2512-18202-sophia-a-persistent-agent-fram.jpg)
[논문 리뷰] Sophia: A Persistent Agent Framework of Artificial Life
The development of LLMs has elevated AI agents from task-specific tools to long-lived, decision-making entities. Yet, most architectures remain static and reactive, tethered to manually defined, narro...
![[논문 리뷰] Distributional AGI Safety](/assets/images/blog/20260103-paper-2512-16856-distributional-agi-safety.jpg)
[논문 리뷰] Distributional AGI Safety
AI safety and alignment research has predominantly been focused on methods for safeguarding individual AI systems, resting on the assumption of an eventual emergence of a monolithic Artificial General...
![[논문 리뷰] LLaDA2.0: Scaling Up Diffusion Language Models to 100B](/assets/images/blog/20260103-paper-2512-15745-llada2-0-scaling-up-diffusion-.jpg)
[논문 리뷰] LLaDA2.0: Scaling Up Diffusion Language Models to 100B
This paper presents LLaDA2.0 -- a tuple of discrete diffusion large language models (dLLM) scaling up to 100B total parameters through systematic conversion from auto-regressive (AR) models -- establi...
![[논문 리뷰] ReFusion: A Diffusion Large Language Model with Parallel Autoregressive Decoding](/assets/images/blog/20260103-paper-2512-13586-refusion-a-diffusion-large-lan.jpg)
[논문 리뷰] ReFusion: A Diffusion Large Language Model with Parallel Autoregressive Decoding
Autoregressive models (ARMs) are hindered by slow sequential inference. While masked diffusion models (MDMs) offer a parallel alternative, they suffer from critical drawbacks: high computational overh...
![[논문 리뷰] Seedance 1.5 pro: A Native Audio-Visual Joint Generation Foundation Model](/assets/images/blog/20260103-paper-2512-13507-seedance-1-5-pro-a-native-audi.jpg)
[논문 리뷰] Seedance 1.5 pro: A Native Audio-Visual Joint Generation Foundation Model
Recent strides in video generation have paved the way for unified audio-visual generation. In this work, we present Seedance 1.5 pro, a foundational model engineered specifically for native, joint aud...
![[논문 리뷰] An Anatomy of Vision-Language-Action Models: From Modules to Milestones and Challenges](/assets/images/blog/20260103-paper-2512-11362-an-anatomy-of-vision-language-.jpg)
[논문 리뷰] An Anatomy of Vision-Language-Action Models: From Modules to Milestones and Challenges
Vision-Language-Action (VLA) models are driving a revolution in robotics, enabling machines to understand instructions and interact with the physical world. This field is exploding with new models and...
![[논문 리뷰] Z-Image: An Efficient Image Generation Foundation Model with Single-Stream Diffusion Transformer](/assets/images/blog/20260103-paper-2511-22699-z-image-an-efficient-image-gen.jpg)
[논문 리뷰] Z-Image: An Efficient Image Generation Foundation Model with Single-Stream Diffusion Transformer
The landscape of high-performance image generation models is currently dominated by proprietary systems, such as Nano Banana Pro and Seedream 4.0. Leading open-source alternatives, including Qwen-Imag...
![[논문 리뷰] ToolOrchestra: Elevating Intelligence via Efficient Model and Tool Orchestration](/assets/images/blog/20260103-paper-2511-21689-toolorchestra-elevating-intell.jpg)
[논문 리뷰] ToolOrchestra: Elevating Intelligence via Efficient Model and Tool Orchestration
Large language models are powerful generalists, yet solving deep and complex problems such as those of the Humanity's Last Exam (HLE) remains both conceptually challenging and computationally expensiv...
![[논문 리뷰] Matrix: Peer-to-Peer Multi-Agent Synthetic Data Generation Framework](/assets/images/blog/20260103-paper-2511-21686-matrix-peer-to-peer-multi-agen.jpg)
[논문 리뷰] Matrix: Peer-to-Peer Multi-Agent Synthetic Data Generation Framework
Synthetic data has become increasingly important for training large language models, especially when real data is scarce, expensive, or privacy-sensitive. Many such generation tasks require coordinate...
![[논문 리뷰] Flow Map Distillation Without Data](/assets/images/blog/20260103-paper-2511-19428-flow-map-distillation-without-.jpg)
[논문 리뷰] Flow Map Distillation Without Data
State-of-the-art flow models achieve remarkable quality but require slow, iterative sampling. To accelerate this, flow maps can be distilled from pre-trained teachers, a procedure that conventionally ...
![[논문 리뷰] Chain-of-Visual-Thought: Teaching VLMs to See and Think Better with Continuous Visual Tokens](/assets/images/blog/20260103-paper-2511-19418-chain-of-visual-thought-teachi.jpg)
[논문 리뷰] Chain-of-Visual-Thought: Teaching VLMs to See and Think Better with Continuous Visual Tokens
Vision-Language Models (VLMs) excel at reasoning in linguistic space but struggle with perceptual understanding that requires dense visual perception, e.g., spatial reasoning and geometric awareness. ...
![[논문 리뷰] CLaRa: Bridging Retrieval and Generation with Continuous Latent Reasoning](/assets/images/blog/20260103-paper-2511-18659-clara-bridging-retrieval-and-g.jpg)
[논문 리뷰] CLaRa: Bridging Retrieval and Generation with Continuous Latent Reasoning
Retrieval-augmented generation (RAG) enhances large language models (LLMs) with external knowledge but still suffers from long contexts and disjoint retrieval-generation optimization. In this work, we...
![[논문 리뷰] OmniScientist: Toward a Co-evolving Ecosystem of Human and AI Scientists](/assets/images/blog/20260103-paper-2511-16931-omniscientist-toward-a-co-evol.jpg)
[논문 리뷰] OmniScientist: Toward a Co-evolving Ecosystem of Human and AI Scientists
With the rapid development of Large Language Models (LLMs), AI agents have demonstrated increasing proficiency in scientific tasks, ranging from hypothesis generation and experimental design to manusc...
![[논문 리뷰] Evolution Strategies at the Hyperscale](/assets/images/blog/20260103-paper-2511-16652-evolution-strategies-at-the-hy.jpg)
[논문 리뷰] Evolution Strategies at the Hyperscale
We introduce Evolution Guided General Optimization via Low-rank Learning (EGGROLL), an evolution strategies (ES) algorithm designed to scale backprop-free optimization to large population sizes for mo...
![[논문 리뷰] A Primer on Quantum Machine Learning](/assets/images/blog/20260103-paper-2511-15969-a-primer-on-quantum-machine-le.jpg)
[논문 리뷰] A Primer on Quantum Machine Learning
Quantum machine learning (QML) is a computational paradigm that seeks to apply quantum-mechanical resources to solve learning problems. As such, the goal of this framework is to leverage quantum proce...
![[논문 리뷰] Fine-Tuned LLMs Know They Don't Know: A Parameter-Efficient Approach to Recovering Honesty](/assets/images/blog/20260103-paper-2511-12991-fine-tuned-llms-know-they-don-.jpg)
[논문 리뷰] Fine-Tuned LLMs Know They Don't Know: A Parameter-Efficient Approach to Recovering Honesty
The honesty of Large Language Models (LLMs) is increasingly important for safe deployment in high-stakes domains. However, this crucial trait is severely undermined by supervised fine-tuning (SFT), a ...
![[논문 리뷰] Black-Box On-Policy Distillation of Large Language Models](/assets/images/blog/20260103-paper-2511-10643-black-box-on-policy-distillati.jpg)
[논문 리뷰] Black-Box On-Policy Distillation of Large Language Models
Black-box distillation creates student large language models (LLMs) by learning from a proprietary teacher model's text outputs alone, without access to its internal logits or parameters. In this work...
![[논문 리뷰] DoPE: Denoising Rotary Position Embedding](/assets/images/blog/20260103-paper-2511-09146-dope-denoising-rotary-position.jpg)
[논문 리뷰] DoPE: Denoising Rotary Position Embedding
Rotary Position Embedding (RoPE) in Transformer models has inherent limits that weaken length extrapolation. We reinterpret the attention map with positional encoding as a noisy feature map, and propo...
![[논문 리뷰] Think-at-Hard: Selective Latent Iterations to Improve Reasoning Language Models](/assets/images/blog/20260103-paper-2511-08577-think-at-hard-selective-latent.jpg)
[논문 리뷰] Think-at-Hard: Selective Latent Iterations to Improve Reasoning Language Models
Improving reasoning capabilities of Large Language Models (LLMs), especially under parameter constraints, is crucial for real-world applications. Prior work proposes recurrent transformers, which allo...
![[논문 리뷰] Tiny Model, Big Logic: Diversity-Driven Optimization Elicits Large-Model Reasoning Ability in VibeThinker-1.5B](/assets/images/blog/20260103-paper-2511-06221-tiny-model-big-logic-diversity.jpg)
[논문 리뷰] Tiny Model, Big Logic: Diversity-Driven Optimization Elicits Large-Model Reasoning Ability in VibeThinker-1.5B
Challenging the prevailing consensus that small models inherently lack robust reasoning, this report introduces VibeThinker-1.5B, a 1.5B-parameter dense model developed via our Spectrum-to-Signal Prin...
![[논문 리뷰] World Simulation with Video Foundation Models for Physical AI](/assets/images/blog/20260103-paper-2511-00062-world-simulation-with-video-fo.jpg)
[논문 리뷰] World Simulation with Video Foundation Models for Physical AI
We introduce [Cosmos-Predict2.5], the latest generation of the Cosmos World Foundation Models for Physical AI. Built on a flow-based architecture, [Cosmos-Predict2.5] unifies Text2World, Image2World, ...
![[논문 리뷰] The Era of Agentic Organization: Learning to Organize with Language Models](/assets/images/blog/20260103-paper-2510-26658-the-era-of-agentic-organizatio.jpg)
[논문 리뷰] The Era of Agentic Organization: Learning to Organize with Language Models
We envision a new era of AI, termed agentic organization, where agents solve complex problems by working collaboratively and concurrently, enabling outcomes beyond individual intelligence. To realize ...
![[논문 리뷰] Compute as Teacher: Turning Inference Compute Into Reference-Free Supervision](/assets/images/blog/20260103-paper-2509-14234-compute-as-teacher-turning-inf.jpg)
[논문 리뷰] Compute as Teacher: Turning Inference Compute Into Reference-Free Supervision
Where do learning signals come from when there is no ground truth in post-training? We propose turning exploration into supervision through Compute as Teacher (CaT), which converts the model's own exp...
![[논문 리뷰] Decoupling the "What" and "Where" With Polar Coordinate Positional Embeddings](/assets/images/blog/20260103-paper-2509-10534-decoupling-the-what-and-where-.jpg)
[논문 리뷰] Decoupling the "What" and "Where" With Polar Coordinate Positional Embeddings
The attention mechanism in a Transformer architecture matches key to query based on both content -- the what -- and position in a sequence -- the where. We present an analysis indicating that what and...
![[논문 리뷰] Emergent Hierarchical Reasoning in LLMs through Reinforcement Learning](/assets/images/blog/20260103-paper-2509-03646-emergent-hierarchical-reasonin.jpg)
[논문 리뷰] Emergent Hierarchical Reasoning in LLMs through Reinforcement Learning
Reinforcement Learning (RL) has proven highly effective at enhancing the complex reasoning abilities of Large Language Models (LLMs), yet underlying mechanisms driving this success remain largely opaq...
![[논문 리뷰] From AI for Science to Agentic Science: A Survey on Autonomous Scientific Discovery](/assets/images/blog/20260103-paper-2508-14111-from-ai-for-science-to-agentic.jpg)
[논문 리뷰] From AI for Science to Agentic Science: A Survey on Autonomous Scientific Discovery
Artificial intelligence (AI) is reshaping scientific discovery, evolving from specialized computational tools into autonomous research partners. We position Agentic Science as a pivotal stage within t...
![[논문 리뷰] Gold-medalist Performance in Solving Olympiad Geometry with AlphaGeometry2](/assets/images/blog/20260103-paper-2502-03544-gold-medalist-performance-in-s.jpg)
[논문 리뷰] Gold-medalist Performance in Solving Olympiad Geometry with AlphaGeometry2
We present AlphaGeometry2 (AG2), a significantly improved version of AlphaGeometry introduced in (Trinh et al., 2024), which has now surpassed an average gold medalist in solving Olympiad geometry pro...
![[논문 리뷰] Nested Learning: The Illusion of Deep Learning Architecture](/assets/images/blog/20260102-paper-url-pdf-nested-learning-the-illusion-o.jpg)
[논문 리뷰] Nested Learning: The Illusion of Deep Learning Architecture
Over the last decades, developing more powerful neural architectures and simultaneously designing optimization algorithms to effectively train them have been the core of research efforts to enhance th...
![[논문 리뷰] Act2Goal: From World Model To General Goal-conditioned Policy](/assets/images/blog/20260102-paper-2512-23541-act2goal-from-world-model-to-g.jpg)
[논문 리뷰] Act2Goal: From World Model To General Goal-conditioned Policy
Specifying robotic manipulation tasks in a manner that is both expressive and precise remains a central challenge. While visual goals provide a compact and unambiguous task specification, existing goa...
![[논문 리뷰] Pruning as a Game: Equilibrium-Driven Sparsification of Neural Networks](/assets/images/blog/20260102-paper-2512-22106-pruning-as-a-game-equilibrium-.jpg)
[논문 리뷰] Pruning as a Game: Equilibrium-Driven Sparsification of Neural Networks
Neural network pruning is widely used to reduce model size and computational cost. Yet, most existing methods treat sparsity as an externally imposed constraint, enforced through heuristic importance ...
![[논문 리뷰] AgentEvolver: Towards Efficient Self-Evolving Agent System](/assets/images/blog/20260102-paper-2511-10395-agentevolver-towards-efficient.jpg)
[논문 리뷰] AgentEvolver: Towards Efficient Self-Evolving Agent System
Autonomous agents powered by large language models (LLMs) have the potential to significantly enhance human productivity by reasoning, using tools, and executing complex tasks in diverse environments....
![[논문 리뷰] TiDAR: Think in Diffusion, Talk in Autoregression](/assets/images/blog/20260102-paper-2511-08923-tidar-think-in-diffusion-talk-.jpg)
[논문 리뷰] TiDAR: Think in Diffusion, Talk in Autoregression
Diffusion language models hold the promise of fast parallel generation, while autoregressive (AR) models typically excel in quality due to their causal structure aligning naturally with language model...
![[논문 리뷰] AlphaResearch: Accelerating New Algorithm Discovery with Language Models](/assets/images/blog/20260102-paper-2511-08522-alpharesearch-accelerating-new.jpg)
[논문 리뷰] AlphaResearch: Accelerating New Algorithm Discovery with Language Models
Large language models have made significant progress in complex but easy-to-verify problems, yet they still struggle with discovering the unknown. In this paper, we present extbf{AlphaResearch}, an ...
![[논문 리뷰] Attention and Compression is all you need for Controllably Efficient Language Models](/assets/images/blog/20260102-paper-2511-05313-attention-and-compression-is-a.jpg)
[논문 리뷰] Attention and Compression is all you need for Controllably Efficient Language Models
The quadratic cost of attention in transformers motivated the development of efficient approaches: namely sparse and sliding window attention, convolutions and linear attention. Although these approac...
![[논문 리뷰] The Curved Spacetime of Transformer Architectures](/assets/images/blog/20260102-paper-2511-03060-the-curved-spacetime-of-transf.jpg)
[논문 리뷰] The Curved Spacetime of Transformer Architectures
We present a geometric framework for understanding Transformer-based language models, drawing an explicit analogy to General Relativity. Queries and keys induce an effective metric on representation s...
![[논문 리뷰] No-Human in the Loop: Agentic Evaluation at Scale for Recommendation](/assets/images/blog/20260102-paper-2511-03051-no-human-in-the-loop-agentic-e.jpg)
[논문 리뷰] No-Human in the Loop: Agentic Evaluation at Scale for Recommendation
Evaluating large language models (LLMs) as judges is increasingly critical for building scalable and trustworthy evaluation pipelines. We present ScalingEval, a large-scale benchmarking study that sys...
![[논문 리뷰] ThinkMorph: Emergent Properties in Multimodal Interleaved Chain-of-Thought Reasoning](/assets/images/blog/20260102-paper-2510-27492-thinkmorph-emergent-properties.jpg)
[논문 리뷰] ThinkMorph: Emergent Properties in Multimodal Interleaved Chain-of-Thought Reasoning
Multimodal reasoning requires iterative coordination between language and vision, yet it remains unclear what constitutes a meaningful interleaved chain of thought. We posit that text and image though...
![[논문 리뷰] Context Engineering 2.0: The Context of Context Engineering](/assets/images/blog/20260102-paper-2510-26493-context-engineering-2-0-the-co.jpg)
[논문 리뷰] Context Engineering 2.0: The Context of Context Engineering
Karl Marx once wrote that ``the human essence is the ensemble of social relations'', suggesting that individuals are not isolated entities but are fundamentally shaped by their interactions with other...
![[논문 리뷰] Chain-of-Thought Hijacking](/assets/images/blog/20260102-paper-2510-26418-chain-of-thought-hijacking.jpg)
[논문 리뷰] Chain-of-Thought Hijacking
Large reasoning models (LRMs) achieve higher task performance with more inference-time computation, and prior works suggest this scaled reasoning may also strengthen safety by improving refusal. Yet w...
![[논문 리뷰] GAP: Graph-Based Agent Planning with Parallel Tool Use and Reinforcement Learning](/assets/images/blog/20260102-paper-2510-25320-gap-graph-based-agent-planning.jpg)
[논문 리뷰] GAP: Graph-Based Agent Planning with Parallel Tool Use and Reinforcement Learning
Autonomous agents powered by large language models (LLMs) have shown impressive capabilities in tool manipulation for complex task-solving. However, existing paradigms such as ReAct rely on sequential...
![[논문 리뷰] Reasoning with Sampling: Your Base Model is Smarter Than You Think](/assets/images/blog/20260102-paper-2510-14901-reasoning-with-sampling-your-b.jpg)
[논문 리뷰] Reasoning with Sampling: Your Base Model is Smarter Than You Think
Frontier reasoning models have exhibited incredible capabilities across a wide array of disciplines, driven by posttraining large language models (LLMs) with reinforcement learning (RL). However, desp...
![[논문 리뷰] Cache-to-Cache: Direct Semantic Communication Between Large Language Models](/assets/images/blog/20260102-paper-2510-03215-cache-to-cache-direct-semantic.jpg)
[논문 리뷰] Cache-to-Cache: Direct Semantic Communication Between Large Language Models
Multi-LLM systems harness the complementary strengths of diverse Large Language Models, achieving performance and efficiency gains unattainable by a single model. In existing designs, LLMs communicate...
![[논문 리뷰] Plan-and-Act: Improving Planning of Agents for Long-Horizon Tasks](/assets/images/blog/20260102-paper-2503-09572-plan-and-act-improving-plannin.jpg)
[논문 리뷰] Plan-and-Act: Improving Planning of Agents for Long-Horizon Tasks
Large language models (LLMs) have shown remarkable advancements in enabling language agents to tackle simple tasks. However, applying them for complex, multi-step, long-horizon tasks remains a challen...
테스트의 중요성과 구현 방법
소프트웨어 개발 과정에서 "테스트"라는 단어를 듣지 않은 개발자는 아마 없을 것입니다. 그만큼 테스트는 소프트웨어의 신뢰성과 안정성을 보장하는 데 필수적인 역할을 합니다. 특히 데이터 과학과 인공지능 분야에서는 결과의 정확성과 모델의 성능을 검증하기 위해 테스트가 더욱 중요합니다. 이번 포스트에서는 테스트의 중요성을 살펴보고, 테스트를 어떻게 효율적으로 구...
![[논문 리뷰] mHC: Manifold-Constrained Hyper-Connections](/assets/images/blog/20260101-paper-2512-24880-mhc-manifold-constrained-hyper.jpg)
[논문 리뷰] mHC: Manifold-Constrained Hyper-Connections
Recently, studies exemplified by Hyper-Connections (HC) have extended the ubiquitous residual connection paradigm established over the past decade by expanding the residual stream width and diversifyi...
![[논문 리뷰] Real Deep Research for AI, Robotics and Beyond](/assets/images/blog/20260101-paper-2510-20809-real-deep-research-for-ai-robo.jpg)
[논문 리뷰] Real Deep Research for AI, Robotics and Beyond
With the rapid growth of research in AI and robotics now producing over 10,000 papers annually it has become increasingly difficult for researchers to stay up to date. Fast evolving trends, the rise o...
![[논문 리뷰] The Free Transformer](/assets/images/blog/20260101-paper-2510-17558-the-free-transformer.jpg)
[논문 리뷰] The Free Transformer
We propose an extension of the decoder Transformer that conditions its generative process on random latent variables which are learned without supervision thanks to a variational procedure. Experiment...
![[논문 리뷰] Training-Free Group Relative Policy Optimization](/assets/images/blog/20260101-paper-2510-08191-training-free-group-relative-p.jpg)
[논문 리뷰] Training-Free Group Relative Policy Optimization
Recent advances in Large Language Model (LLM) agents have demonstrated their promising general capabilities. However, their performance in specialized real-world domains often degrades due to challeng...
![[논문 리뷰] ReasoningBank: Scaling Agent Self-Evolving with Reasoning Memory](/assets/images/blog/20260101-paper-2509-25140-reasoningbank-scaling-agent-se.jpg)
[논문 리뷰] ReasoningBank: Scaling Agent Self-Evolving with Reasoning Memory
With the growing adoption of large language model agents in persistent real-world roles, they naturally encounter continuous streams of tasks. A key limitation, however, is their failure to learn from...
![[논문 리뷰] Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality](/assets/images/blog/20260101-paper-2405-21060-transformers-are-ssms-generali.jpg)
[논문 리뷰] Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality
While Transformers have been the main architecture behind deep learning's success in language modeling, state-space models (SSMs) such as Mamba have recently been shown to match or outperform Transfor...
![[논문 리뷰] Mamba: Linear-Time Sequence Modeling with Selective State Spaces](/assets/images/blog/20260101-paper-2312-00752-mamba-linear-time-sequence-mod.jpg)
[논문 리뷰] Mamba: Linear-Time Sequence Modeling with Selective State Spaces
Foundation models, now powering most of the exciting applications in deep learning, are almost universally based on the Transformer architecture and its core attention module. Many subquadratic-time a...
![[논문 리뷰] Vector database management systems: Fundamental concepts, use-cases, and current challenges](/assets/images/blog/20260101-paper-2309-11322-vector-database-management-sys.jpg)
[논문 리뷰] Vector database management systems: Fundamental concepts, use-cases, and current challenges
Vector database management systems have emerged as an important component in modern data management, driven by the growing importance for the need to computationally describe rich data such as texts, ...
데이터 증강을 통한 모델 성능 향상 기법
인공지능(AI)와 머신러닝(ML) 분야에서 데이터는 가장 중요한 자산입니다. 충분한 양의 고품질 데이터를 확보하는 것은 모델의 성능을 결정짓는 중요한 요소입니다. 그러나 현실에서는 데이터가 부족하거나, 데이터 수집에 많은 비용과 시간이 소요되는 경우가 자주 발생합니다. 이러한 문제를 해결하기 위해 데이터 증강(Data Augmentation) 기법이 주목받...
RAG 시스템 구축: 검색 증강 생성의 원리와 구현
오늘날 인공지능(AI)이 발전하면서 자연어 처리(NLP) 분야에서도 다양한 혁신이 일어나고 있습니다. 그 중 하나가 바로 RAG(Retrieval-Augmented Generation, 검색 증강 생성) 시스템입니다. RAG는 검색과 생성 두 가지 프로세스를 결합하여 보다 정확하고 풍부한 정보를 제공하는 데 중점을 둡니다. 전통적인 NLP 모델이 대규모 데...
PyTorch 텐서 기초
딥러닝(Deep Learning)은 현대 인공지능(AI) 기술의 중심에 서 있습니다. 이 중에서도 PyTorch는 연구자와 개발자들 사이에서 인기가 높은 프레임워크로, 그 유연성과 직관적인 인터페이스 덕분에 널리 사용되고 있습니다. PyTorch를 제대로 이해하기 위해서는 텐서(Tensor)라는 기본 단위를 확실히 이해하는 것이 중요합니다. 텐서는 데이터를...
![[논문 리뷰] QLoRA: Efficient Finetuning of Quantized LLMs](/assets/images/blog/20251231-paper-2305-14314-qlora-efficient-finetuning-of-.jpg)
[논문 리뷰] QLoRA: Efficient Finetuning of Quantized LLMs
We present QLoRA, an efficient finetuning approach that reduces memory usage enough to finetune a 65B parameter model on a single 48GB GPU while preserving full 16-bit finetuning task performance. QLo...
![[논문 리뷰] BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](/assets/images/blog/20251231-paper-1810-04805-bert-pre-training-of-deep-bidi.jpg)
[논문 리뷰] BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
We introduce a new language representation model called BERT, which stands for Bidirectional Encoder Representations from Transformers. Unlike recent language representation models, BERT is designed t...
MLOps 입문: 머신러닝 운영 파이프라인 구축
머신러닝(ML)은 오늘날 많은 산업에서 혁신을 주도하고 있습니다. 그러나 머신러닝 모델을 성공적으로 개발하는 것만으로는 충분하지 않습니다. 모델을 실제 운영 환경에 배포하고 모니터링하며 지속적으로 개선하기 위해서는 특별한 노력이 필요합니다. 이 과정에서 MLOps가 중요한 역할을 합니다....
LLM 파인튜닝: LoRA와 QLoRA 기법
대규모 언어 모델(LLM, Large Language Models)은 자연어 처리(NLP, Natural Language Processing) 분야에서 혁신을 이끌어왔습니다. 이러한 모델들은 방대한 양의 데이터를 활용하여 다양한 언어적 과제를 수행할 수 있습니다. 하지만 LLM은 막대한 컴퓨팅 자원을 요구하며, 특정 작업에 맞게 모델을 조정하는 파인튜닝 과...
Diffusion 모델: 이미지 생성 AI의 원리와 활용
인공지능(AI) 기술은 최근 몇 년 동안 급격히 발전해 왔으며, 특히 이미지 생성 분야에서 큰 주목을 받고 있습니다. 이러한 발전의 중심에는 'Diffusion 모델'이라는 강력한 기술이 자리잡고 있습니다. Diffusion 모델은 복잡한 패턴을 학습하고, 현실감 넘치는 이미지를 생성하는 데 탁월한 성능을 보이며, 다양한 산업 분야에서 응용되고 있습니다....
딥러닝 기초
딥러닝(Deep Learning)은 인공지능(AI) 분야에서 가장 주목받는 기술 중 하나로, 여러 분야에서 혁신적인 변화를 이끌고 있습니다. 이미지를 인식하거나 자연어를 처리하는 것과 같은 복잡한 문제를 해결하는 데 있어 딥러닝의 역할은 날이 갈수록 커지고 있습니다. 이 글에서는 딥러닝의 기본 개념을 이해하고, 실제로 어떻게 구현할 수 있는지를 살펴보겠습니...
자연어처리에서의 Word Embedding
자연어처리(NLP, Natural Language Processing) 분야에서 Word Embedding은 필수적인 개념 중 하나입니다. 이는 컴퓨터가 인간의 언어를 이해하고 처리할 수 있도록 돕는 중요한 기술입니다. 이번 블로그 포스트에서는 Word Embedding의 기본 개념과 주요 기법들을 살펴보고, Python 코드를 통해 이를 실습해보는 시간을...
트랜스포머 Attention 메커니즘의 이해
최근 몇 년간 자연어 처리(Natural Language Processing, NLP) 분야에서는 혁신적인 변화가 일어났습니다. 그 중심에는 단연 트랜스포머(Transformer) 모델이 자리 잡고 있습니다. 트랜스포머는 다양한 NLP 작업에서 뛰어난 성능을 보이며, 언어 모델, 번역, 요약, 질의응답 등 여러 응용 분야에서 사용됩니다. 이러한 트랜스포머의...
강화학습 완벽 가이드: 이론부터 실전까지
강화학습(Reinforcement Learning)은 인공지능(AI) 분야에서 기계가 스스로 학습하고 결정할 수 있는 능력을 부여하는 중요한 기술입니다. 이 방법론은 로봇 제어, 게임 플레이, 자율 주행 자동차 등 다양한 분야에서 혁신적인 결과를 보여주고 있습니다. 본 가이드에서는 강화학습의 역사, 이론적 배경, 주요 알고리즘, 그리고 실제 구현까지 심층적으로 다룹니다.
![[논문 리뷰] Attention Is All You Need](/assets/images/blog/20251230-paper-1706-03762-attention-is-all-you-need.jpg)
[논문 리뷰] Attention Is All You Need
Transformer 아키텍처를 최초로 제안한 획기적인 논문. RNN과 CNN을 완전히 배제하고 Attention 메커니즘만으로 시퀀스 모델링의 새로운 패러다임을 제시하여, BERT, GPT 등 현대 자연어처리의 기반을 마련했습니다.
Graph Neural Networks 기초
최근 들어 인공지능(AI)과 머신러닝(ML)이 다양한 분야에서 혁신을 이루고 있습니다. 그중에서도 그래프 뉴럴 네트워크(Graph Neural Networks, GNN)는 복잡한 구조적 데이터를 효율적으로 처리할 수 있는 강력한 도구로 주목받고 있습니다. 그래프 데이터는 소셜 네트워크, 추천 시스템, 분자 구조 등 다양한 실세계 문제에서 자연스럽게 발생하며...
CNN 이미지 분류: 딥러닝의 핵심 기술
이미지 분류는 컴퓨터 비전 분야에서 가장 중요한 문제 중 하나로, 자율주행차의 장애물 인식, 의료 영상 분석, 소셜 미디어의 이미지 태그링 등 다양한 분야에 응용되고 있습니다. 특히, CNN(Convolutional Neural Networks, 합성곱 신경망)은 이미지 데이터 처리에서 탁월한 성능을 보여주며, 딥러닝 혁신의 중심에 서 있습니다. 이번 블로...
BERT 완벽 가이드: 자연어처리의 혁명적 모델
BERT(Bidirectional Encoder Representations from Transformers)는 2018년 구글이 발표한 혁신적인 자연어처리 모델입니다. 양방향 문맥 이해를 통해 NLP 분야에 새로운 패러다임을 제시한 BERT의 아키텍처, 학습 방법, 실전 활용법까지 상세히 알아봅니다.
SuanLab 블로그에 오신 것을 환영합니다
데이터 과학, 인공지능, 딥러닝에 관한 다양한 이야기를 공유하는 SuanLab 블로그입니다.