PyTorchKR Papers

“Code Generation”

총 10,186편
태그·기간 등 상세 필터

카테고리 탐색

arXiv:2607.27578v1cs.AI

무엇이 프롬프트를 그래프로 만드는가: 프롬프트 그래프 엔지니어링을 위한 필요충분조건

What makes prompts a graph: necessary and sufficient conditions for prompt graph engineering

프롬프트는 이미 오래전에 개별적인 문자열에 머무르기를 멈췄습니다. 실제 시스템에서는 하나의 모델 호출이 다른 호출로 이어지고, 검색이 생성과 교차되며, 라우터가 분기하고, 애그리게이터가 병렬 결과를 병합합니다. 실무에서는 이를 하나로 묶기 위한 단일 구조인 그래프로 수렴했습니다. LangGraph, DSPy, Prompt Flow와 같은 프레임워크는 이를 공개적으로 드러내며, 연구 시스템에서는 이미 이를 자동으로 최적화하고 있습니다. 그러나 관련 용어는 이러한 발전을 따라가지 못하고 있습니다. '그래프'는 하나의 샘플링 전략 내부의 추론 토폴로지, 다중 에이전트 대화, 오케스트레이션 아티팩트 등을 다양하게 지칭하는 반면, '프롬프트 엔지니어링'은 여전히 훌륭한 단일 문자열을 작성하는 것만을 연상시킵니다. 현재 결여된 것은 프롬프트를 명시적이고 실행 가능하며 개선 가능한 그래프의 노드로 다루는 기준 정의입니다. 본 논문에서는 주요 회색 문헌(grey literature)으로 보완되고 영구 식별자를 갖춘 출처들에 대한 개념 분석을 통해 이러한 정의를 구축합니다. 데이터플로우 그래프 및 빌드 시스템에서 시작하여 프롬프트 체이닝과 사고 토폴로지(체인, 트리, 그래프)를 거쳐, 아티팩트로서 컴파일 및 최적화되는 그래프에 이르기까지 이 아이디어의 계보를 재구성합니다. 그런 다음 프롬프트 그래프 엔지니어링의 구성적 정의를 제안하고, 4가지 조건(명시적 구조, 구조와 프롬프트 내용의 분리, 실행 가능한 시맨틱, 일급 엔지니어링 아티팩트로서의 그래프)을 명시하며, 이를 포함 및 배제 테스트로 조작화(operationalize)합니다. 또한 6가지 인접 개념과의 경계를 설정하고 6개의 실제 시스템(LangGraph, DSPy, Prompt Flow, AutoGen, CrewAI, Claude Code 하위 에이전트)에 이 테스트를 적용하여 일관되게 포함 및 배제됨을 확인합니다. 마지막으로 4가지 설계 긴장 축을 중심으로 정리된 연구 과제를 제시하며 마무리합니다. 본 연구의 기여는 업계에서 명확한 명칭 없이 이미 일상적으로 수행하고 있는 실무에 대한 조작적 정의와 공유 어휘를 제공하는 것입니다.

arXiv:2609.03871v1cs.AI

Bioinfoysis Technical Report

Large language model agents have shown promise in bioinformatics, but most existing systems focus primarily on producing final answers, treating planning, tool use, and code execution as transient interactions. This design is poorly suited to long-horizon bioinformatics tasks, where conclusions must remain connected to the data, computations, and intermediate evidence that support them. We introduce \textbf{Bioinfoysis}, a multi-agent harness that represents each request as a persistent, artifact-grounded analysis run. Bioinfoysis combines global planning with step-wise, evidence-driven replanning: the planner maintains an executable checklist and revises pending steps using structured handoffs returned after each worker execution. These handoffs bind intermediate results to their responsible agent, checklist step, and plan generation, preventing stale evidence from being silently reused after replanning. A controlled runtime validates generated scripts, tables, and figures before they are used in downstream analysis or reporting, while role-specific context, persistent memory, and governed bioinformatics skills support reliable execution over long analysis trajectories. We evaluate Bioinfoysis on BixBench and two question-answering tracks of LAB-Bench 2. On BixBench, Bioinfoysis achieves state-of-the-art accuracy of 82.4\%. Across four underlying language models, Bioinfoysis increases average accuracy from 27.81\% to 64.13\% on SeqQA2 and from 3.13\% to 31.25\% on DbQA2. These results demonstrate that reliable bioinformatics automation depends not only on model capability, but also on the harness that governs planning, execution, memory, and evidence flow. We hope that the emergence of Bioinfoysis will play a driving and leading role in the development of the bioinformatics community. Our demo website can be seen in https://report.bioinfoysis.com/.

arXiv:1912.06112공식 코드 · Papers with Code

Unified Generative Adversarial Networks for Controllable Image-to-Image Translation

We propose a unified Generative Adversarial Network (GAN) for controllable image-to-image translation, i.e., transferring an image from a source to a target domain guided by controllable structures. In addition to conditioning on a reference image, we show how the model can generate images conditioned on controllable structures, e.g., class labels, object keypoints, human skeletons, and scene semantic maps. The proposed model consists of a single generator and a discriminator taking a conditional image and the target controllable structure as input. In this way, the conditional image can provide appearance information and the controllable structure can provide the structure information for generating the target result. Moreover, our model learns the image-to-image mapping through three novel losses, i.e., color loss, controllable structure guided cycle-consistency loss, and controllable structure guided self-content preserving loss. Also, we present the Fr\'echet ResNet Distance (FRD) to evaluate the quality of the generated images. Experiments on two challenging image translation tasks, i.e., hand gesture-to-gesture translation and cross-view image translation, show that our model generates convincing results, and significantly outperforms other state-of-the-art methods on both tasks. Meanwhile, the proposed framework is a unified solution, thus it can be applied to solving other controllable structure guided image translation tasks such as landmark guided facial expression translation and keypoint guided person image generation. To the best of our knowledge, we are the first to make one GAN framework work on all such controllable structure guided image translation tasks. Code is available at https://github.com/Ha0Tang/GestureGAN.

arXiv:2007.04774공식 코드 · Papers with Code

Automated Chest CT Image Segmentation of COVID-19 Lung Infection based on 3D U-Net

The coronavirus disease 2019 (COVID-19) affects billions of lives around the world and has a significant impact on public healthcare. Due to rising skepticism towards the sensitivity of RT-PCR as screening method, medical imaging like computed tomography offers great potential as alternative. For this reason, automated image segmentation is highly desired as clinical decision support for quantitative assessment and disease monitoring. However, publicly available COVID-19 imaging data is limited which leads to overfitting of traditional approaches. To address this problem, we propose an innovative automated segmentation pipeline for COVID-19 infected regions, which is able to handle small datasets by utilization as variant databases. Our method focuses on on-the-fly generation of unique and random image patches for training by performing several preprocessing methods and exploiting extensive data augmentation. For further reduction of the overfitting risk, we implemented a standard 3D U-Net architecture instead of new or computational complex neural network architectures. Through a 5-fold cross-validation on 20 CT scans of COVID-19 patients, we were able to develop a highly accurate as well as robust segmentation model for lungs and COVID-19 infected regions without overfitting on the limited data. Our method achieved Dice similarity coefficients of 0.956 for lungs and 0.761 for infection. We demonstrated that the proposed method outperforms related approaches, advances the state-of-the-art for COVID-19 segmentation and improves medical image analysis with limited data. The code and model are available under the following link: https://github.com/frankkramer-lab/covid19.MIScnn

arXiv:2008.05570공식 코드 · Papers with Code

PLACE: Proximity Learning of Articulation and Contact in 3D Environments

High fidelity digital 3D environments have been proposed in recent years, however, it remains extremely challenging to automatically equip such environment with realistic human bodies. Existing work utilizes images, depth or semantic maps to represent the scene, and parametric human models to represent 3D bodies. While being straightforward, their generated human-scene interactions are often lack of naturalness and physical plausibility. Our key observation is that humans interact with the world through body-scene contact. To synthesize realistic human-scene interactions, it is essential to effectively represent the physical contact and proximity between the body and the world. To that end, we propose a novel interaction generation method, named PLACE (Proximity Learning of Articulation and Contact in 3D Environments), which explicitly models the proximity between the human body and the 3D scene around it. Specifically, given a set of basis points on a scene mesh, we leverage a conditional variational autoencoder to synthesize the minimum distances from the basis points to the human body surface. The generated proximal relationship exhibits which region of the scene is in contact with the person. Furthermore, based on such synthesized proximity, we are able to effectively obtain expressive 3D human bodies that interact with the 3D scene naturally. Our perceptual study shows that PLACE significantly improves the state-of-the-art method, approaching the realism of real human-scene interaction. We believe our method makes an important step towards the fully automatic synthesis of realistic 3D human bodies in 3D scenes. The code and model are available for research at https://sanweiliti.github.io/PLACE/PLACE.html.

arXiv:2012.03352공식 코드 · Papers with Code

An Uncertainty-Driven GCN Refinement Strategy for Organ Segmentation

Organ segmentation in CT volumes is an important pre-processing step in many computer assisted intervention and diagnosis methods. In recent years, convolutional neural networks have dominated the state of the art in this task. However, since this problem presents a challenging environment due to high variability in the organ's shape and similarity between tissues, the generation of false negative and false positive regions in the output segmentation is a common issue. Recent works have shown that the uncertainty analysis of the model can provide us with useful information about potential errors in the segmentation. In this context, we proposed a segmentation refinement method based on uncertainty analysis and graph convolutional networks. We employ the uncertainty levels of the convolutional network in a particular input volume to formulate a semi-supervised graph learning problem that is solved by training a graph convolutional network. To test our method we refine the initial output of a 2D U-Net. We validate our framework with the NIH pancreas dataset and the spleen dataset of the medical segmentation decathlon. We show that our method outperforms the state-of-the-art CRF refinement method by improving the dice score by 1% for the pancreas and 2% for spleen, with respect to the original U-Net's prediction. Finally, we perform a sensitivity analysis on the parameters of our proposal and discuss the applicability to other CNN architectures, the results, and current limitations of the model for future work in this research direction. For reproducibility purposes, we make our code publicly available at https://github.com/rodsom22/gcn_refinement.

arXiv:2202.12595공식 코드 · Papers with Code

Evolutionary scheduling of university activities based on consumption forecasts to minimise electricity costs

This paper presents a solution to a predict then optimise problem which goal is to reduce the electricity cost of a university campus. The proposed methodology combines a multi-dimensional time series forecast and a novel approach to large-scale optimization. Gradient-boosting method is applied to forecast both generation and consumption time-series of the Monash university campus for the month of November 2020. For the consumption forecasts we employ log transformation to model trend and stabilize variance. Additional seasonality and trend features are added to the model inputs when applicable. The forecasts obtained are used as the base load for the schedule optimisation of university activities and battery usage. The goal of the optimisation is to minimize the electricity cost consisting of the price of electricity and the peak electricity tariff both altered by the load from class activities and battery use as well as the penalty of not scheduling some optional activities. The schedule of the class activities is obtained through evolutionary optimisation using the covariance matrix adaptation evolution strategy and the genetic algorithm. This schedule is then improved through local search by testing possible times for each activity one-by-one. The battery schedule is formulated as a mixed-integer programming problem and solved by the Gurobi solver. This method obtains the second lowest cost when evaluated against 6 other methods presented at an IEEE competition that all used mixed-integer programming and the Gurobi solver to schedule both the activities and the battery use. The code and data used for the paper are publicly available.

arXiv:2301.11387공식 코드 · Papers with Code

Universal Domain Adaptation for Remote Sensing Image Scene Classification

The domain adaptation (DA) approaches available to date are usually not well suited for practical DA scenarios of remote sensing image classification, since these methods (such as unsupervised DA) rely on rich prior knowledge about the relationship between label sets of source and target domains, and source data are often not accessible due to privacy or confidentiality issues. To this end, we propose a practical universal domain adaptation setting for remote sensing image scene classification that requires no prior knowledge on the label sets. Furthermore, a novel universal domain adaptation method without source data is proposed for cases when the source data is unavailable. The architecture of the model is divided into two parts: the source data generation stage and the model adaptation stage. The first stage estimates the conditional distribution of source data from the pre-trained model using the knowledge of class-separability in the source domain and then synthesizes the source data. With this synthetic source data in hand, it becomes a universal DA task to classify a target sample correctly if it belongs to any category in the source label set, or mark it as ``unknown" otherwise. In the second stage, a novel transferable weight that distinguishes the shared and private label sets in each domain promotes the adaptation in the automatically discovered shared label set and recognizes the ``unknown'' samples successfully. Empirical results show that the proposed model is effective and practical for remote sensing image scene classification, regardless of whether the source data is available or not. The code is available at https://github.com/zhu-xlab/UniDA.

arXiv:2308.14852공식 코드 · Papers with Code

SynthDistill: Face Recognition with Knowledge Distillation from Synthetic Data

State-of-the-art face recognition networks are often computationally expensive and cannot be used for mobile applications. Training lightweight face recognition models also requires large identity-labeled datasets. Meanwhile, there are privacy and ethical concerns with collecting and using large face recognition datasets. While generating synthetic datasets for training face recognition models is an alternative option, it is challenging to generate synthetic data with sufficient intra-class variations. In addition, there is still a considerable gap between the performance of models trained on real and synthetic data. In this paper, we propose a new framework (named SynthDistill) to train lightweight face recognition models by distilling the knowledge of a pretrained teacher face recognition model using synthetic data. We use a pretrained face generator network to generate synthetic face images and use the synthesized images to learn a lightweight student network. We use synthetic face images without identity labels, mitigating the problems in the intra-class variation generation of synthetic datasets. Instead, we propose a novel dynamic sampling strategy from the intermediate latent space of the face generator network to include new variations of the challenging images while further exploring new face images in the training batch. The results on five different face recognition datasets demonstrate the superiority of our lightweight model compared to models trained on previous synthetic datasets, achieving a verification accuracy of 99.52% on the LFW dataset with a lightweight network. The results also show that our proposed framework significantly reduces the gap between training with real and synthetic data. The source code for replicating the experiments is publicly released.

arXiv:2312.01886공식 코드 · Papers with Code

InstructTA: Instruction-Tuned Targeted Attack for Large Vision-Language Models

Large vision-language models (LVLMs) have demonstrated their incredible capability in image understanding and response generation. However, this rich visual interaction also makes LVLMs vulnerable to adversarial examples. In this paper, we formulate a novel and practical targeted attack scenario that the adversary can only know the vision encoder of the victim LVLM, without the knowledge of its prompts (which are often proprietary for service providers and not publicly available) and its underlying large language model (LLM). This practical setting poses challenges to the cross-prompt and cross-model transferability of targeted adversarial attack, which aims to confuse the LVLM to output a response that is semantically similar to the attacker's chosen target text. To this end, we propose an instruction-tuned targeted attack (dubbed \textsc{InstructTA}) to deliver the targeted adversarial attack on LVLMs with high transferability. Initially, we utilize a public text-to-image generative model to "reverse" the target response into a target image, and employ GPT-4 to infer a reasonable instruction $\boldsymbol{p}^\prime$ from the target response. We then form a local surrogate model (sharing the same vision encoder with the victim LVLM) to extract instruction-aware features of an adversarial image example and the target image, and minimize the distance between these two features to optimize the adversarial example. To further improve the transferability with instruction tuning, we augment the instruction $\boldsymbol{p}^\prime$ with instructions paraphrased from GPT-4. Extensive experiments demonstrate the superiority of our proposed method in targeted attack performance and transferability. The code is available at https://github.com/xunguangwang/InstructTA.

arXiv:2401.01578공식 코드 · Papers with Code

Context-Guided Spatio-Temporal Video Grounding

Spatio-temporal video grounding (or STVG) task aims at locating a spatio-temporal tube for a specific instance given a text query. Despite advancements, current methods easily suffer the distractors or heavy object appearance variations in videos due to insufficient object information from the text, leading to degradation. Addressing this, we propose a novel framework, context-guided STVG (CG-STVG), which mines discriminative instance context for object in videos and applies it as a supplementary guidance for target localization. The key of CG-STVG lies in two specially designed modules, including instance context generation (ICG), which focuses on discovering visual context information (in both appearance and motion) of the instance, and instance context refinement (ICR), which aims to improve the instance context from ICG by eliminating irrelevant or even harmful information from the context. During grounding, ICG, together with ICR, are deployed at each decoding stage of a Transformer architecture for instance context learning. Particularly, instance context learned from one decoding stage is fed to the next stage, and leveraged as a guidance containing rich and discriminative object feature to enhance the target-awareness in decoding feature, which conversely benefits generating better new instance context for improving localization finally. Compared to existing methods, CG-STVG enjoys object information in text query and guidance from mined instance visual context for more accurate target localization. In our experiments on three benchmarks, including HCSTVG-v1/-v2 and VidSTG, CG-STVG sets new state-of-the-arts in m_tIoU and m_vIoU on all of them, showing its efficacy. The code will be released at https://github.com/HengLan/CGSTVG.

arXiv:2402.18668공식 코드 · Papers with Code

Simple linear attention language models balance the recall-throughput tradeoff

Recent work has shown that attention-based language models excel at recall, the ability to ground generations in tokens previously seen in context. However, the efficiency of attention-based models is bottle-necked during inference by the KV-cache's aggressive memory consumption. In this work, we explore whether we can improve language model efficiency (e.g. by reducing memory consumption) without compromising on recall. By applying experiments and theory to a broad set of architectures, we identify a key tradeoff between a model's state size and recall ability. We show that efficient alternatives to attention (e.g. H3, Mamba, RWKV) maintain a fixed-size recurrent state, but struggle at recall. We propose BASED a simple architecture combining linear and sliding window attention. By varying BASED window size and linear attention feature dimension, we can dial the state size and traverse the pareto frontier of the recall-memory tradeoff curve, recovering the full quality of attention on one end and the small state size of attention-alternatives on the other. We train language models up to 1.3b parameters and show that BASED matches the strongest sub-quadratic models (e.g. Mamba) in perplexity and outperforms them on real-world recall-intensive tasks by 6.22 accuracy points. Implementations of linear attention are often less efficient than optimized standard attention implementations. To make BASED competitive, we develop IO-aware algorithms that enable 24x higher throughput on language generation than FlashAttention-2, when generating 1024 tokens using 1.3b parameter models. Code for this work is provided at: https://github.com/HazyResearch/based.

arXiv:2404.03868공식 코드 · Papers with Code

Extract, Define, Canonicalize: An LLM-based Framework for Knowledge Graph Construction

In this work, we are interested in automated methods for knowledge graph creation (KGC) from input text. Progress on large language models (LLMs) has prompted a series of recent works applying them to KGC, e.g., via zero/few-shot prompting. Despite successes on small domain-specific datasets, these models face difficulties scaling up to text common in many real-world applications. A principal issue is that, in prior methods, the KG schema has to be included in the LLM prompt to generate valid triplets; larger and more complex schemas easily exceed the LLMs' context window length. Furthermore, there are scenarios where a fixed pre-defined schema is not available and we would like the method to construct a high-quality KG with a succinct self-generated schema. To address these problems, we propose a three-phase framework named Extract-Define-Canonicalize (EDC): open information extraction followed by schema definition and post-hoc canonicalization. EDC is flexible in that it can be applied to settings where a pre-defined target schema is available and when it is not; in the latter case, it constructs a schema automatically and applies self-canonicalization. To further improve performance, we introduce a trained component that retrieves schema elements relevant to the input text; this improves the LLMs' extraction performance in a retrieval-augmented generation-like manner. We demonstrate on three KGC benchmarks that EDC is able to extract high-quality triplets without any parameter tuning and with significantly larger schemas compared to prior works. Code for EDC is available at https://github.com/clear-nus/edc.

arXiv:2407.00696공식 코드 · Papers with Code

Graph in Graph Neural Network

Existing Graph Neural Networks (GNNs) are limited to process graphs each of whose vertices is represented by a vector or a single value, limited their representing capability to describe complex objects. In this paper, we propose the first GNN (called Graph in Graph Neural (GIG) Network) which can process graph-style data (called GIG sample) whose vertices are further represented by graphs. Given a set of graphs or a data sample whose components can be represented by a set of graphs (called multi-graph data sample), our GIG network starts with a GIG sample generation (GSG) module which encodes the input as a \textbf{GIG sample}, where each GIG vertex includes a graph. Then, a set of GIG hidden layers are stacked, with each consisting of: (1) a GIG vertex-level updating (GVU) module that individually updates the graph in every GIG vertex based on its internal information; and (2) a global-level GIG sample updating (GGU) module that updates graphs in all GIG vertices based on their relationships, making the updated GIG vertices become global context-aware. This way, both internal cues within the graph contained in each GIG vertex and the relationships among GIG vertices could be utilized for down-stream tasks. Experimental results demonstrate that our GIG network generalizes well for not only various generic graph analysis tasks but also real-world multi-graph data analysis (e.g., human skeleton video-based action recognition), which achieved the new state-of-the-art results on 13 out of 14 evaluated datasets. Our code is publicly available at https://github.com/wangjs96/Graph-in-Graph-Neural-Network.

arXiv:2412.17707공식 코드 · Papers with Code

SMAC-Hard: Enabling Mixed Opponent Strategy Script and Self-play on SMAC

The availability of challenging simulation environments is pivotal for advancing the field of Multi-Agent Reinforcement Learning (MARL). In cooperative MARL settings, the StarCraft Multi-Agent Challenge (SMAC) has gained prominence as a benchmark for algorithms following centralized training with decentralized execution paradigm. However, with continual advancements in SMAC, many algorithms now exhibit near-optimal performance, complicating the evaluation of their true effectiveness. To alleviate this problem, in this work, we highlight a critical issue: the default opponent policy in these environments lacks sufficient diversity, leading MARL algorithms to overfit and exploit unintended vulnerabilities rather than learning robust strategies. To overcome these limitations, we propose SMAC-HARD, a novel benchmark designed to enhance training robustness and evaluation comprehensiveness. SMAC-HARD supports customizable opponent strategies, randomization of adversarial policies, and interfaces for MARL self-play, enabling agents to generalize to varying opponent behaviors and improve model stability. Furthermore, we introduce a black-box testing framework wherein agents are trained without exposure to the edited opponent scripts but are tested against these scripts to evaluate the policy coverage and adaptability of MARL algorithms. We conduct extensive evaluations of widely used and state-of-the-art algorithms on SMAC-HARD, revealing the substantial challenges posed by edited and mixed strategy opponents. Additionally, the black-box strategy tests illustrate the difficulty of transferring learned policies to unseen adversaries. We envision SMAC-HARD as a critical step toward benchmarking the next generation of MARL algorithms, fostering progress in self-play methods for multi-agent systems. Our code is available at https://github.com/devindeng94/smac-hard.

arXiv:2610.40966v1cs.CL

더 나은 검색기, 더 나쁜 레이블: 불완전한 관련성 판단이 대화형 RAG의 응답 보류 평가를 왜곡하는 방식

Better Retrievers, Worse Labels: How Incomplete Relevance Judgments Distort Abstention Evaluation in Conversational RAG

MTRAG와 RETECO 같은 대화형 RAG 벤치마크는 검색과 생성을 종단 간으로 평가한다. 각 시스템은 자체 검색 결과를 바탕으로 답하므로, 응답 보류(abstention)는 검색된 각 컨텍스트가 충분한지에 비추어 채점되어야 한다. 편의상 사용되는 지름길은 관련성 판단에서 충분성을 곧바로 읽어 내어, 정답 문서(gold document)가 없는 컨텍스트를 답변 불가능한 것으로 취급한다. 우리는 공식 베이스라인을 그대로 다시 실행하고, RECOR(RETECO 트랙 2)에서 네 가지 검색기(nDCG@10이 .18에서 .59)와 세 가지 Qwen3 리더로 이 지름길을 검증한다. 두 개의 오픈 LLM 판정기와 네 개의 소형 오픈 모델로 구성된 패널이 내용에 따라 각 컨텍스트에 레이블을 부여하며, 다른 계열의 모델이 250개 컨텍스트를 블라인드로 감사하고 조정 절차를 거친다. 검색기가 강할수록 정답 문서가 없는데도 답을 포함하는 컨텍스트의 비율은 커지고(28%에서 72%), 지름길의 정밀도는 낮아진다(.68에서 .10, 감사 기준으로는 .70에서 .21). 리더도 이에 부합한다. Qwen3-4B는 이러한 허위 미탐(false miss)의 62–67%에 올바르게 답하는 반면, 불충분하다고 판정된 정답 문서 없는 컨텍스트에서는 23–30%, 문서가 전혀 없을 때는 1%에 답한다. 대부분의 허위 미탐에서 답은 같은 질문에 대해 채굴된 하드 네거티브, 즉 다른 턴의 정답 문서 또는 정답 페이지의 인접 청크에서 나온다. 판단이 주석자에게서 나온 MTRAG에서는 정밀도가 .72에서 .50으로 떨어진다. 지름길을 사용하면 답변 가능성 AUROC가 최대 .19까지 과소 추정되며, 최고의 충분성 탐지기는 문서를 전혀 읽지 않는 검색 점수처럼 보이게 된다. 우리는 코드, 실행 결과, 레이블을 공개하며, 정답 문서의 존재 여부가 아니라 내용 레이블에 따라 거절을 채점할 것을 권고한다.

arXiv:2605.13481v5cs.CL

PersonalAI 2.0: 개인화 LLM 에이전트를 위한 계획 메커니즘 기반 지식 그래프 탐색/검색 강화

PersonalAI 2.0: Enhancing knowledge graph traversal/retrieval with planning mechanism for Personalized LLM Agents

외부 지식 그래프(KG)의 통합을 통해 LLM 기반 시스템을 향상시키도록 설계된 새로운 프레임워크인 PersonalAI 2.0(PAI-2)을 소개한다. 제안하는 접근법은 동적인 다단계 파이프라인을 도입하여 기존 그래프 검색 증강 생성(GraphRAG) 방법의 주요 한계를 해결한다. PAI-2 설계의 핵심은 추출된 개체, 일치하는 그래프 정점, 생성된 단서 질의에 의해 안내되는 적응적이고 반복적인 정보 탐색을 수행하는 능력이다. 다섯 개의 벤치마크(Natural Questions, TriviaQA, HotpotQA, 2WikiMultihopQA, MuSiQue)에서 수행한 평가는 기준 방법(LightRAG, RAPTOR, HippoRAG 2, PAI-1)에 비해 생성된 답변의 사실적 정확성이 향상됨을 보여 준다. PAI-2는 2WikiMultihopQA와 MuSiQue 벤치마크에서 LLM-as-a-Judge 기준 평균 9퍼센트포인트(pp)의 향상을 달성하고, Qwen2.5 7B 및 multilingual-e5-large를 사용한 TriviaQA와 HotpotQA 벤치마크에서는 HippoRAG 2와 비슷한 정확도를 달성하여, 환각 비율을 줄이고 재현율을 높이는 효과를 보인다. 검색 계획 강화 메커니즘을 활성화하면 비활성화한 경우에 비해 다섯 개 벤치마크 전반에서 LLM-as-a-Judge 기준 평균 18 pp의 향상을 얻음을 보인다. 또한 절제 연구에서 PAI-2가 MINE-1 벤치마크에서 Qwen2.5 7B 및 multilingual-e5-large로 89%의 정보 보존 점수를 얻어 우수한 결과를 달성함을 확인한다. 종합하면, 이러한 결과는 PAI-2가 문맥 인지 지식 표현과 추론 능력을 필요로 하는 개인화 AI 응용을 위한 재사용 가능한 구성 요소로 기능할 잠재력을 보여 준다. PAI-2의 소스 코드는 https://github.com/Dzigen/PersonalAI 에서 확인할 수 있다.

arXiv:2607.05682v2cs.AI

FirstResearch: 대규모 언어 모델 과학적 발견 에이전트를 위한 감사 가능한 연구 질문 형성

FirstResearch: Auditable Question Formation for LLM Scientific Discovery Agents

FirstResearch는 LLM 기반 과학적 발견 에이전트가 제안하는 연구 질문의 감사 가능성(auditability)을 높이기 위해, 기본 정의·가정·메커니즘·반증 가능한 가설·실패 업데이트 규칙을 명시적으로 기록하는 Research Question Certificate를 핵심 산출물로 삼는 프레임워크를 제안한다. 10개 주제 벤치마크에서 AI co-scientist, Agent Laboratory, AI Scientist-v2에서 착안한 기준선 대비 DeepSeek 및 Gemini-2.5-Flash 평가 모두에서 최고 점수(4.86–4.90/5)를 기록하였으며, 인증서 제거 시 점수가 1점 미만으로 급락해 명시적 도출 제약 조건의 효과를 뒷받침한다.

arXiv:2605.12768v3stat.ML

ISOMORPH: 시뮬레이션, 데이터셋 생성 및 예측 벤치마크를 위한 공급망 디지털 트윈

ISOMORPH: A Supply Chain Digital Twin for Simulation, Dataset Generation, and Forecasting Benchmarks

공개 시계열 예측(TSF) 벤치마크는 소매, 에너지, 기상, 교통 등을 다루고 있으나 공급망 물류 분야는 여전히 부족한 실정이다. 본 논문에서는 해석 가능하고 사용자가 설정할 수 있는 매개변수와 모듈형 토폴로지, 수요 및 제어 규칙을 갖춘 다단계 물류 네트워크의 최초 공개 디지털 트윈인 ISOMORPH를 소개한다. 이 시뮬레이터는 이산 시간에서 방향성 라우팅 그래프를 진행시킨다. 즉, 수요는 재고에서 충족되거나 백로그(미처리 주문)로 기록되며 네트워크 전반에 걸쳐 보충을 촉발한다. 상태는 재고, 미결 주문, 운송 중인 선적, 평활화된 수요 추정치를 추적하여 다루기 쉬운 상태 공간에서 마르코프 역학을 생성한다. 공개된 데이터는 실증적으로 일관된 규모로 채찍 효과(bullwhip effect)를 재현하며, 세 가지 보존 법칙은 시뮬레이터 확장을 위한 검증 도구를 제공한다. 우리는 두 가지 카탈로그 규모($C=50$ 및 $C=200$)의 데이터셋을 공개하며, $C=50$에서 33개 롤아웃 시나리오 라이브러리를 함께 제공한다. 이러한 데이터셋은 분산 증폭, 연쇄적 병목 현상, 체제 전환(regime shift), 공유된 거시적 충격을 통한 채널 간 결합 등 기존의 고정된 TSF 벤치마크에서는 대체로 찾아볼 수 없는 역학을 보여준다. 세 가지 도메인 내 학습된 베이스라인(ARIMA, ETS, PatchTST) 대비 세 가지 파운데이션 모델(Chronos, Moirai, TimesFM)의 제로샷(zero-shot) 평가는 수요, 백로그, 주문 충족률(fill rate), 엣지 활용도의 네 가지 타깃에 걸쳐 진행된다. ETTh1, Electricity, Weather와의 비교는 ISOMORPH가 표준 실제 TSF 벤치마크와 다른 예측 체제를 도입함을 보여주며, 상호 보완적이고 재생성 가능한 물류 도메인 벤치마크로서의 입지를 확립한다. 동일한 페어링은 시나리오 구성 전반에 걸쳐 예측 신뢰 구간을 생성하여 매개변수 불확실성으로부터 순방향 불확실성 정량화(UQ)를 제공하고, 파운데이션 모델이 디지털 트윈 기반 UQ를 위한 빠른 대리 모델(surrogate)로 기능할 수 있음을 입증한다. 코드(MIT): https://github.com/tuhinsahai/ISOMORPH. 인터랙티브 데모: https://huggingface.co/spaces/HyeminGu/ISOMORPH-demo.

arXiv:2109.04075공식 코드 · Papers with Code

Self Supervision to Distillation for Long-Tailed Visual Recognition

Deep learning has achieved remarkable progress for visual recognition on large-scale balanced datasets but still performs poorly on real-world long-tailed data. Previous methods often adopt class re-balanced training strategies to effectively alleviate the imbalance issue, but might be a risk of over-fitting tail classes. The recent decoupling method overcomes over-fitting issues by using a multi-stage training scheme, yet, it is still incapable of capturing tail class information in the feature learning stage. In this paper, we show that soft label can serve as a powerful solution to incorporate label correlation into a multi-stage training scheme for long-tailed recognition. The intrinsic relation between classes embodied by soft labels turns out to be helpful for long-tailed recognition by transferring knowledge from head to tail classes. Specifically, we propose a conceptually simple yet particularly effective multi-stage training scheme, termed as Self Supervised to Distillation (SSD). This scheme is composed of two parts. First, we introduce a self-distillation framework for long-tailed recognition, which can mine the label relation automatically. Second, we present a new distillation label generation module guided by self-supervision. The distilled labels integrate information from both label and data domains that can model long-tailed distribution effectively. We conduct extensive experiments and our method achieves the state-of-the-art results on three long-tailed recognition benchmarks: ImageNet-LT, CIFAR100-LT and iNaturalist 2018. Our SSD outperforms the strong LWS baseline by from $2.7\%$ to $4.5\%$ on various datasets. The code is available at https://github.com/MCG-NJU/SSD-LT.