https://suanlab.com/blog 이수안 교수의 데이터 사이언스 블로그 - AI, 딥러닝, 머신러닝, 자연어처리, 컴퓨터 비전 ko Sun, 12 Jul 2026 11:53:51 GMT https://suanlab.com/assets/images/logo.png https://suanlab.com/blog suan@suanlab.com (이수안) suan@suanlab.com (이수안) Copyright 2026 SuanLab. All rights reserved. 60 https://suanlab.com/blog/20260626-paper-2512-14391-repo-language-models-with-cont https://suanlab.com/blog/20260626-paper-2512-14391-repo-language-models-with-cont In-context learning is fundamental to modern Large Language Models (LLMs); however, prevailing architectures impose a rigid and fixed contextual structure by assigning linear or constant positional in... Fri, 26 Jun 2026 00:00:00 GMT Paper Review Paper Review cs.LG cs.AI cs.CL suan@suanlab.com (이수안) https://suanlab.com/blog/20260624-paper-2606-09498-self-harness-harnesses-that-im https://suanlab.com/blog/20260624-paper-2606-09498-self-harness-harnesses-that-im The performance of LLM-based agents is jointly shaped by their base models and the harnesses that mediate their interaction with the environment. Because different models exhibit distinct behaviors, e... Wed, 24 Jun 2026 00:00:00 GMT Paper Review Paper Review cs.CL cs.CL suan@suanlab.com (이수안) https://suanlab.com/blog/20260624-paper-2605-26379-when-does-lejepa-learn-a-world https://suanlab.com/blog/20260624-paper-2605-26379-when-does-lejepa-learn-a-world A representation that scrambles the true degrees of freedom of the world cannot support reliable planning or compositional generalization. We prove that LeJEPA (alignment plus Gaussian regularization)... Wed, 24 Jun 2026 00:00:00 GMT Paper Review Paper Review stat.ML cs.LG stat.ML suan@suanlab.com (이수안) https://suanlab.com/blog/20260624-paper-2605-23904-skillopt-executive-strategy-fo https://suanlab.com/blog/20260624-paper-2605-23904-skillopt-executive-strategy-fo Agent skills today are hand-crafted, generated one-shot, or evolved through loosely controlled self-revision, none of which behaves like a deep-learning optimizer for the skill, and none of which reli... Wed, 24 Jun 2026 00:00:00 GMT Paper Review Paper Review cs.AI cs.CL cs.AI suan@suanlab.com (이수안) https://suanlab.com/blog/20260622-paper-2602-24281-memory-caching-rnns-with-growi https://suanlab.com/blog/20260622-paper-2602-24281-memory-caching-rnns-with-growi Transformers have been established as the de-facto backbones for most recent advances in sequence modeling, mainly due to their growing memory capacity that scales with the context length. While plaus... Mon, 22 Jun 2026 00:00:00 GMT Paper Review Paper Review cs.LG cs.AI cs.LG suan@suanlab.com (이수안) https://suanlab.com/blog/20260613-paper-2606-12683-from-agi-to-asi https://suanlab.com/blog/20260613-paper-2606-12683-from-agi-to-asi Over the last decade, building human-level artificial general intelligence has moved from far-fetched speculation to being a concrete next-decade target for many of the largest AI organisations. Achie... Sat, 13 Jun 2026 00:00:00 GMT Paper Review Paper Review cs.AI cs.CY cs.LG suan@suanlab.com (이수안) https://suanlab.com/blog/20260613-paper-2606-09659-end-to-end-context-compression https://suanlab.com/blog/20260613-paper-2606-09659-end-to-end-context-compression Long-context language model inference is bottlenecked by memory, as the KV cache grows with context length. Recent techniques to compress the KV cache fall short: they either degrade model quality sub... Sat, 13 Jun 2026 00:00:00 GMT Paper Review Paper Review cs.CL cs.AI cs.LG suan@suanlab.com (이수안) https://suanlab.com/blog/20260607-paper-2606-05400-leanmarathon-toward-reliable-a https://suanlab.com/blog/20260607-paper-2606-05400-leanmarathon-toward-reliable-a Long-horizon autoformalization of research mathematics fails not only at hard lemmas, but at scale: statements drift, dependencies tangle, context decays, and local repairs corrupt distant work. We pr... Sun, 07 Jun 2026 00:00:00 GMT Paper Review Paper Review cs.AI cs.CL cs.LG suan@suanlab.com (이수안) https://suanlab.com/blog/20260607-paper-2606-01444-self-revising-discovery-system https://suanlab.com/blog/20260607-paper-2606-01444-self-revising-discovery-system Scientific discovery is not only answer generation but revision of the representational regime in which evidence, artifacts, operations, and verifiers are typed. We develop a category-theoretic accoun... Sun, 07 Jun 2026 00:00:00 GMT Paper Review Paper Review cs.AI cond-mat.mtrl-sci cs.CL suan@suanlab.com (이수안) https://suanlab.com/blog/20260607-paper-2602-24281-memory-caching-rnns-with-growi https://suanlab.com/blog/20260607-paper-2602-24281-memory-caching-rnns-with-growi Transformers have been established as the de-facto backbones for most recent advances in sequence modeling, mainly due to their growing memory capacity that scales with the context length. While plaus... Sun, 07 Jun 2026 00:00:00 GMT Paper Review Paper Review cs.LG cs.AI cs.LG suan@suanlab.com (이수안) https://suanlab.com/blog/20260607-paper-2505-17117-from-tokens-to-thoughts-how-ll https://suanlab.com/blog/20260607-paper-2505-17117-from-tokens-to-thoughts-how-ll Humans organize knowledge into compact conceptual categories that balance compression with semantic richness. Large Language Models (LLMs) exhibit impressive linguistic abilities, but whether they nav... Sun, 07 Jun 2026 00:00:00 GMT Paper Review Paper Review cs.CL cs.AI cs.IT suan@suanlab.com (이수안) https://suanlab.com/blog/20260603-paper-2605-20670-lt2-linear-time-looped-transfo https://suanlab.com/blog/20260603-paper-2605-20670-lt2-linear-time-looped-transfo Looped Transformers (LT) have emerged as a powerful architecture by iterating their layers multiple times before decoding the final token. However, pairing them with full attention retains quadratic c... Wed, 03 Jun 2026 00:00:00 GMT Paper Review Paper Review cs.LG cs.LG suan@suanlab.com (이수안) https://suanlab.com/blog/20260603-paper-2605-01428-hallucinations-undermine-trust https://suanlab.com/blog/20260603-paper-2605-01428-hallucinations-undermine-trust Despite significant strides in factual reliability, errors -- often termed hallucinations -- remain a major concern for generative AI, especially as LLMs are increasingly expected to be helpful in mor... Wed, 03 Jun 2026 00:00:00 GMT Paper Review Paper Review cs.CL cs.CL suan@suanlab.com (이수안) https://suanlab.com/blog/20260603-paper-2602-23643-ai-must-embrace-specialization https://suanlab.com/blog/20260603-paper-2602-23643-ai-must-embrace-specialization Everyone from AI executives and researchers to doomsayers, politicians, and activists is talking about Artificial General Intelligence (AGI). Yet, they often don't seem to agree on its exact definitio... Wed, 03 Jun 2026 00:00:00 GMT Paper Review Paper Review cs.AI cs.AI suan@suanlab.com (이수안) https://suanlab.com/blog/20260602-paper-2605-28655-autoscientists-self-organizing https://suanlab.com/blog/20260602-paper-2605-28655-autoscientists-self-organizing Scientific research proceeds through iterative cycles of hypothesis generation, experiment design, execution, and revision. AI agents can automate parts of this process, but existing approaches typica... Tue, 02 Jun 2026 00:00:00 GMT Paper Review Paper Review cs.AI cs.AI suan@suanlab.com (이수안) https://suanlab.com/blog/20260602-paper-2605-18661-ai-for-auto-research-roadmap-a https://suanlab.com/blog/20260602-paper-2605-18661-ai-for-auto-research-roadmap-a AI-assisted research is crossing a threshold: fully automated systems can now generate research papers for as little as $15, while long-horizon agents can execute experiments, draft manuscripts, and s... Tue, 02 Jun 2026 00:00:00 GMT Paper Review Paper Review cs.AI cs.AI suan@suanlab.com (이수안) https://suanlab.com/blog/20260602-paper-2605-10878-neural-weight-norm-kolmogorov- https://suanlab.com/blog/20260602-paper-2605-10878-neural-weight-norm-kolmogorov- Why does weight decay work? We prove that, in any fixed-precision regime, the smallest weight norm of a looped neural network outputting a binary string equals the Kolmogorov complexity of that string... Tue, 02 Jun 2026 00:00:00 GMT Paper Review Paper Review cs.LG cs.IT cs.LG suan@suanlab.com (이수안) https://suanlab.com/blog/20260526-paper-2605-26099-language-models-need-sleep https://suanlab.com/blog/20260526-paper-2605-26099-language-models-need-sleep Transformer-based large language models are increasingly used for long-horizon tasks; however, their attention mechanism scales poorly with context length. To handle this, we study a sleep-like consol... Tue, 26 May 2026 00:00:00 GMT Paper Review Paper Review cs.CL cs.AI cs.CL suan@suanlab.com (이수안) https://suanlab.com/blog/20260526-paper-2605-23904-skillopt-executive-strategy-fo https://suanlab.com/blog/20260526-paper-2605-23904-skillopt-executive-strategy-fo Agent skills today are hand-crafted, generated one-shot, or evolved through loosely controlled self-revision, none of which behaves like a deep-learning optimizer for the skill, and none of which reli... Tue, 26 May 2026 00:00:00 GMT Paper Review Paper Review cs.AI cs.CL cs.AI suan@suanlab.com (이수안) https://suanlab.com/blog/20260524-paper-2605-22821-tokenisation-via-convex-relaxa https://suanlab.com/blog/20260524-paper-2605-22821-tokenisation-via-convex-relaxa Tokenisation is an integral part of the current NLP pipeline. Current tokenisation algorithms such as BPE and Unigram are greedy algorithms -- they make locally optimal decisions without considering t... Sun, 24 May 2026 00:00:00 GMT Paper Review Paper Review cs.CL cs.LG cs.CL suan@suanlab.com (이수안) https://suanlab.com/blog/20260524-paper-2605-22763-advancing-mathematics-research https://suanlab.com/blog/20260524-paper-2605-22763-advancing-mathematics-research Large language models (LLMs) increasingly excel at mathematical reasoning, but their unreliability limits their utility in mathematics research. A mitigation is using LLMs to generate formal proofs in... Sun, 24 May 2026 00:00:00 GMT Paper Review Paper Review cs.AI cs.AI suan@suanlab.com (이수안) https://suanlab.com/blog/20260524-paper-2605-21488-equilibrium-reasoners-learning https://suanlab.com/blog/20260524-paper-2605-21488-equilibrium-reasoners-learning Scaling test-time compute by iteratively updating a latent state has emerged as a powerful paradigm for reasoning. Yet the internal mechanisms that enable these iterative models to generalize beyond m... Sun, 24 May 2026 00:00:00 GMT Paper Review Paper Review cs.LG cs.LG suan@suanlab.com (이수안) https://suanlab.com/blog/20260524-paper-2605-01428-hallucinations-undermine-trust https://suanlab.com/blog/20260524-paper-2605-01428-hallucinations-undermine-trust Despite significant strides in factual reliability, errors -- often termed hallucinations -- remain a major concern for generative AI, especially as LLMs are increasingly expected to be helpful in mor... Sun, 24 May 2026 00:00:00 GMT Paper Review Paper Review cs.CL cs.CL suan@suanlab.com (이수안) https://suanlab.com/blog/20260524-paper-2603-01092-the-alien-space-of-science-sam https://suanlab.com/blog/20260524-paper-2603-01092-the-alien-space-of-science-sam Scientific discovery is constrained not only by what is true, but by what is cognitively available to the researchers currently exploring a field. Many directions are coherent in light of the literatu... Sun, 24 May 2026 00:00:00 GMT Paper Review Paper Review cs.AI cs.LG cs.AI suan@suanlab.com (이수안) https://suanlab.com/blog/20260522-paper-2605-19376-generative-recursive-reasoning https://suanlab.com/blog/20260522-paper-2605-19376-generative-recursive-reasoning How should future neural reasoning systems implement extended computation? Recursive Reasoning Models (RRMs) offer a promising alternative to autoregressive sequence extension by performing iterative ... Fri, 22 May 2026 00:00:00 GMT Paper Review Paper Review cs.AI cs.AI suan@suanlab.com (이수안) https://suanlab.com/blog/20260522-paper-2605-16321-language-game-talking-to-non-h https://suanlab.com/blog/20260522-paper-2605-16321-language-game-talking-to-non-h Language carries thought and coordination among humans but rarely reaches further along the spectrum of diverse intelligence. Yet non-neural systems -- from gene regulatory networks and microbial cons... Fri, 22 May 2026 00:00:00 GMT Paper Review Paper Review cs.LG cs.LG suan@suanlab.com (이수안) https://suanlab.com/blog/20260522-paper-2512-04695-trinity-an-evolved-llm-coordin https://suanlab.com/blog/20260522-paper-2512-04695-trinity-an-evolved-llm-coordin Combining diverse foundation models is promising, but weight-merging is limited by mismatched architectures and closed APIs. Trinity addresses this with a lightweight coordinator that orchestrates col... Fri, 22 May 2026 00:00:00 GMT Paper Review Paper Review cs.LG cs.LG suan@suanlab.com (이수안) https://suanlab.com/blog/20260521-paper-2605-18747-code-as-agent-harness https://suanlab.com/blog/20260521-paper-2605-18747-code-as-agent-harness Recent large language models (LLMs) have demonstrated strong capabilities in understanding and generating code, from competitive programming to repository-level software engineering. In emerging agent... Thu, 21 May 2026 00:00:00 GMT Paper Review Paper Review cs.CL cs.AI cs.CL suan@suanlab.com (이수안) https://suanlab.com/blog/20260518-paper-2603-21687-mirage-the-illusion-of-visual- https://suanlab.com/blog/20260518-paper-2603-21687-mirage-the-illusion-of-visual- Multimodal AI systems have achieved remarkable performance across a broad range of real-world tasks, yet the mechanisms underlying visual-language reasoning remain surprisingly poorly understood. We r... Mon, 18 May 2026 00:00:00 GMT Paper Review Paper Review cs.AI cs.AI suan@suanlab.com (이수안) https://suanlab.com/blog/20260518-paper-2601-21484-ets-energy-guided-test-time-sc https://suanlab.com/blog/20260518-paper-2601-21484-ets-energy-guided-test-time-sc Reinforcement Learning (RL) post-training alignment for language models is effective, but also costly and unstable in practice, owing to its complicated training process. To address this, we propose a... Mon, 18 May 2026 00:00:00 GMT Paper Review Paper Review cs.LG cs.LG suan@suanlab.com (이수안) https://suanlab.com/blog/20260518-paper-2512-16301-adaptation-of-agentic-ai-a-sur https://suanlab.com/blog/20260518-paper-2512-16301-adaptation-of-agentic-ai-a-sur Large language model (LLM) agents are moving beyond prompting alone. ChatGPT marked the rise of general-purpose LLM assistants, DeepSeek showed that on-policy reinforcement learning with verifiable re... Mon, 18 May 2026 00:00:00 GMT Paper Review Paper Review cs.AI cs.CL cs.AI suan@suanlab.com (이수안) https://suanlab.com/blog/20260518-paper-2410-05229-gsm-symbolic-understanding-the https://suanlab.com/blog/20260518-paper-2410-05229-gsm-symbolic-understanding-the Recent advancements in Large Language Models (LLMs) have sparked interest in their formal reasoning capabilities, particularly in mathematics. The GSM8K benchmark is widely used to assess the mathemat... Mon, 18 May 2026 00:00:00 GMT Paper Review Paper Review cs.LG cs.AI cs.LG suan@suanlab.com (이수안) https://suanlab.com/blog/20260516-paper-2510-04476-compressed-convolutional-atten https://suanlab.com/blog/20260516-paper-2510-04476-compressed-convolutional-atten Multi-headed Attention's (MHA) quadratic compute and linearly growing KV-cache make long-context transformers expensive to train and serve. Prior works such as Grouped Query Attention (GQA) and Multi-... Sat, 16 May 2026 00:00:00 GMT Paper Review Paper Review cs.CL cs.AI cs.CL suan@suanlab.com (이수안) https://suanlab.com/blog/20260515-paper-2605-07721-memory-efficient-looped-transf https://suanlab.com/blog/20260515-paper-2605-07721-memory-efficient-looped-transf Recurrent LLM architectures have emerged as a promising approach for improving reasoning, as they enable multi-step computation in the embedding space without generating intermediate tokens. Models su... Fri, 15 May 2026 00:00:00 GMT Paper Review Paper Review cs.CL cs.AI cs.LG suan@suanlab.com (이수안) https://suanlab.com/blog/20260515-paper-2605-06651-ai-co-mathematician-accelerati https://suanlab.com/blog/20260515-paper-2605-06651-ai-co-mathematician-accelerati We introduce the AI co-mathematician, a workbench for mathematicians to interactively leverage AI agents to pursue open-ended research. The AI co-mathematician is optimized to provide holistic support... Fri, 15 May 2026 00:00:00 GMT Paper Review Paper Review cs.AI cs.AI suan@suanlab.com (이수안) https://suanlab.com/blog/20260515-paper-2405-03042-functional-post-clustering-sel https://suanlab.com/blog/20260515-paper-2405-03042-functional-post-clustering-sel In electronic health records (EHR) analysis, clustering patients according to patterns in their data is crucial for uncovering new subtypes of diseases. Existing medical literature often relies on cla... Fri, 15 May 2026 00:00:00 GMT Paper Review Paper Review stat.ME stat.AP stat.CO suan@suanlab.com (이수안) https://suanlab.com/blog/20260515-paper-2404-19737-better-amp-faster-large-langua https://suanlab.com/blog/20260515-paper-2404-19737-better-amp-faster-large-langua Large language models such as GPT and Llama are trained with a next-token prediction loss. In this work, we suggest that training language models to predict multiple future tokens at once results in h... Fri, 15 May 2026 00:00:00 GMT Paper Review Paper Review cs.CL cs.CL suan@suanlab.com (이수안) https://suanlab.com/blog/20260514-paper-2605-12671-all-circuits-lead-to-rome-reth https://suanlab.com/blog/20260514-paper-2605-12671-all-circuits-lead-to-rome-reth In this paper, we present empirical and theoretical evidence against a central but largely implicit assumption in circuit and sheaf discovery (CSD), which we term the Functional Anisotropy Hypothesis:... Thu, 14 May 2026 00:00:00 GMT Paper Review Paper Review cs.CL cs.CL suan@suanlab.com (이수안) https://suanlab.com/blog/20260514-paper-2605-10938-elf-embedded-language-flows https://suanlab.com/blog/20260514-paper-2605-10938-elf-embedded-language-flows Diffusion and flow-based models have become the de facto approaches for generating continuous data, e.g., in domains such as images and videos. Their success has attracted growing interest in applying... Thu, 14 May 2026 00:00:00 GMT Paper Review Paper Review cs.CL cs.AI cs.LG suan@suanlab.com (이수안) https://suanlab.com/blog/20260514-paper-2605-08513-a-single-neuron-is-sufficient- https://suanlab.com/blog/20260514-paper-2605-08513-a-single-neuron-is-sufficient- Safety alignment in language models operates through two mechanistically distinct systems: refusal neurons that gate whether harmful knowledge is expressed, and concept neurons that encode the harmful... Thu, 14 May 2026 00:00:00 GMT Paper Review Paper Review cs.CL cs.AI cs.LG suan@suanlab.com (이수안) https://suanlab.com/blog/20260514-paper-2605-08083-llms-improving-llms-agentic-di https://suanlab.com/blog/20260514-paper-2605-08083-llms-improving-llms-agentic-di Test-time scaling (TTS) has become an effective approach for improving large language model performance by allocating additional computation during inference. However, existing TTS strategies are larg... Thu, 14 May 2026 00:00:00 GMT Paper Review Paper Review cs.CL cs.CL suan@suanlab.com (이수안) https://suanlab.com/blog/20260513-paper-2605-10123-complex-valued-phase-coherent- https://suanlab.com/blog/20260513-paper-2605-10123-complex-valued-phase-coherent- Complex-valued Transformers have largely inherited softmax attention from real-valued architectures. However, row-normalised token competition is not necessarily aligned with phase-preserving computat... Wed, 13 May 2026 00:00:00 GMT Paper Review Paper Review cs.LG cs.LG suan@suanlab.com (이수안) https://suanlab.com/blog/20260513-paper-2605-08044-fast-byte-latent-transformer https://suanlab.com/blog/20260513-paper-2605-08044-fast-byte-latent-transformer Recent byte-level language models (LMs) match the performance of token-level models without relying on subword vocabularies, yet their utility is limited by slow, byte-by-byte autoregressive generatio... Wed, 13 May 2026 00:00:00 GMT Paper Review Paper Review cs.CL cs.AI cs.LG suan@suanlab.com (이수안) https://suanlab.com/blog/20260511-paper-2603-12231-temporal-straightening-for-lat https://suanlab.com/blog/20260511-paper-2603-12231-temporal-straightening-for-lat Learning good representations is essential for latent planning with world models. While pretrained visual encoders produce strong semantic visual features, they are not tailored to planning and contai... Mon, 11 May 2026 00:00:00 GMT Paper Review Paper Review cs.LG cs.LG suan@suanlab.com (이수안) https://suanlab.com/blog/20260509-post https://suanlab.com/blog/20260509-post 최근 몇 년간 인공지능(AI) 분야에서 가장 주목할 만한 발전은 단연 대규모 언어 모델(Large Language Model, LLM)의 부상입니다. OpenAI의 GPT-4, Google의 Gemini, Meta의 Llama 3와 같은 모델들은 자연어 처리(Natural Language Processing, NLP) 분야에 혁신을 가져왔습니다. 이러한 모... Sat, 09 May 2026 00:00:00 GMT General suan@suanlab.com (이수안) https://suanlab.com/blog/20260509-paper-2605-01172-a-theory-of-generalization-in- https://suanlab.com/blog/20260509-paper-2605-01172-a-theory-of-generalization-in- We present a non-asymptotic theory of generalization in deep learning where the empirical neural tangent kernel partitions the output space. In directions corresponding to signal, error dissipates rap... Sat, 09 May 2026 00:00:00 GMT Paper Review Paper Review cs.LG stat.ML cs.LG suan@suanlab.com (이수안) https://suanlab.com/blog/20260509-paper-2603-12231-temporal-straightening-for-lat https://suanlab.com/blog/20260509-paper-2603-12231-temporal-straightening-for-lat Learning good representations is essential for latent planning with world models. While pretrained visual encoders produce strong semantic visual features, they are not tailored to planning and contai... Sat, 09 May 2026 00:00:00 GMT Paper Review Paper Review cs.LG cs.LG suan@suanlab.com (이수안) https://suanlab.com/blog/20260507-paper-2604-27660-from-context-to-skills-can-lan https://suanlab.com/blog/20260507-paper-2604-27660-from-context-to-skills-can-lan Many real-world tasks require language models (LMs) to reason over complex contexts that exceed their parametric knowledge. This calls for context learning, where LMs directly learn relevant knowledge... Thu, 07 May 2026 00:00:00 GMT Paper Review Paper Review cs.AI cs.AI suan@suanlab.com (이수안) https://suanlab.com/blog/20260506-paper-2504-20951-information-gravity-a-field-th https://suanlab.com/blog/20260506-paper-2504-20951-information-gravity-a-field-th We propose a theoretical model called "information gravity" to describe the text generation process in large language models (LLMs). The model uses physical apparatus from field theory and spacetime g... Wed, 06 May 2026 00:00:00 GMT Paper Review Paper Review cs.CL cs.CL suan@suanlab.com (이수안) https://suanlab.com/blog/20260505-human-in-the-loop https://suanlab.com/blog/20260505-human-in-the-loop 자연언어처리(Natural Language Processing, NLP)는 인공지능(AI) 분야에서 가장 빠르게 발전하는 영역 중 하나입니다. 특히 거대 언어 모델(Large Language Models, LLM)의 등장은 인간의 언어를 이해하고 생성하는 AI의 능력을 전례 없는 수준으로 끌어올렸습니다. 하지만 이러한 눈부신 발전에도 불구하고, AI는 여전... Tue, 05 May 2026 00:00:00 GMT General suan@suanlab.com (이수안)