공대생 도전 일지

https://aclanthology.org/2023.findings-emnlp.153/ Chain of Thought with Explicit Evidence Reasoning for Few-shot Relation ExtractionXilai Ma, Jing Li, Min Zhang. Findings of the Association for Computational Linguistics: EMNLP 2023. 2023.aclanthology.org 기존의 FSRE(Few-shot Relation Extraction)은 제한된 학습 샘플을 이용해 두 엔티티간 관계를 예측하는 문제를 다룬다.소수의 주석된 샘플 데이터만을 사용해 관계를 학습하고 예측해야하는 상황으로 메타 학습이나 신경 그래프, In-Con..

https://arxiv.org/abs/2102.09690 Calibrate Before Use: Improving Few-Shot Performance of Language ModelsGPT-3 can perform numerous tasks when provided a natural language prompt that contains a few training examples. We show that this type of few-shot learning can be unstable: the choice of prompt format, training examples, and even the order of the trainingarxiv.org 우리가 Few-Shot을 사용하면 언어 모델은 이에 ..

https://aclanthology.org/2023.findings-emnlp.101/ What Makes Chain-of-Thought Prompting Effective? A Counterfactual StudyAman Madaan, Katherine Hermann, Amir Yazdanbakhsh. Findings of the Association for Computational Linguistics: EMNLP 2023. 2023.aclanthology.org CoT는 LLM의 성능을 높이는데 사용되는 방법입니다.그러나 CoT를 통한 성능 증가의 이유는 아직 밝혀지지 않았고, 패턴, 기호, 잘못된 정보, 조작을 진행하여 다양한 조건에서 테스트를 진행합니다.   이러한 결과를 보면 CoT는 Few..

https://arxiv.org/abs/2210.03493 Automatic Chain of Thought Prompting in Large Language ModelsLarge language models (LLMs) can perform complex reasoning by generating intermediate reasoning steps. Providing these steps for prompting demonstrations is called chain-of-thought (CoT) prompting. CoT prompting has two major paradigms. One leverages a simarxiv.org CoT의 성능은 입증되었지만 CoT의 예제 입력 방식은 상당히 귀찮은..

https://arxiv.org/abs/2305.14325 Improving Factuality and Reasoning in Language Models through Multiagent DebateLarge language models (LLMs) have demonstrated remarkable capabilities in language generation, understanding, and few-shot learning in recent years. An extensive body of work has explored how their performance may be further improved through the tools of parxiv.org Agent 논문입니다! 그 중에서도 ..

https://arxiv.org/abs/2005.14165 Language Models are Few-Shot LearnersRecent work has demonstrated substantial gains on many NLP tasks and benchmarks by pre-training on a large corpus of text followed by fine-tuning on a specific task. While typically task-agnostic in architecture, this method still requires task-specific fiarxiv.orghttps://arxiv.org/abs/2201.11903 Chain-of-Thought Prompting Eli..

https://arxiv.org/abs/2201.11903 Chain-of-Thought Prompting Elicits Reasoning in Large Language ModelsWe explore how generating a chain of thought -- a series of intermediate reasoning steps -- significantly improves the ability of large language models to perform complex reasoning. In particular, we show how such reasoning abilities emerge naturally in suarxiv.org 직전에 봤던 논문의 연장선 같은 느낌입니다.Few-Sh..

https://arxiv.org/abs/2005.14165 Language Models are Few-Shot LearnersRecent work has demonstrated substantial gains on many NLP tasks and benchmarks by pre-training on a large corpus of text followed by fine-tuning on a specific task. While typically task-agnostic in architecture, this method still requires task-specific fiarxiv.orgFew-Shot은 이 그림으로 명확하게 설명이 가능하겠네요 파라미터의 변경 없이 Prompt에 몇 개의 예시만으로..

https://arxiv.org/abs/2303.11366 Reflexion: Language Agents with Verbal Reinforcement LearningLarge language models (LLMs) have been increasingly used to interact with external environments (e.g., games, compilers, APIs) as goal-driven agents. However, it remains challenging for these language agents to quickly and efficiently learn from trial-and-arxiv.org LLM이 지속된 자기 반성을 통해 학습을 진행하여 강화학습과 같이 성..

https://arxiv.org/abs/2501.12948 DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement LearningWe introduce our first-generation reasoning models, DeepSeek-R1-Zero and DeepSeek-R1. DeepSeek-R1-Zero, a model trained via large-scale reinforcement learning (RL) without supervised fine-tuning (SFT) as a preliminary step, demonstrates remarkable reasoninarxiv.org 2025.02.02 - [인공지..