2024년 07월 17일 일일 요약

=============================================

1: 이 글은 소프트웨어 개발자들이 데이터 분석을 위해 DuckDB를 사용해야 하는 세 가지 이유에 대해 설명합니다. 첫째, DuckDB는 널리 이해되는 SQL을 사용하여 다양한 파일 형식을 처리할 수 있습니다. 둘째, 여러 데이터베이스와 파일 형식을 지원하여 교차 마이크로서비스 질문을 쉽게 해결할 수 있습니다. 셋째, DuckDB는 포터블하고 확장 가능하며, 브라우저에서도 실행될 수 있어 다양한 프로그래밍 언어와 환경에서 사용할 수 있습니다.

키워드: DuckDB, SQL, 데이터 분석, 데이터베이스, 포터블

출처: https://medium.com/@mourjo/three-reasons-why-developers-should-use-duckdb-0884c8e9f02a?source=email-5cbf792c976b-1721149920634-digest.reader-7f60cf5620c9-0884c8e9f02a----2-98------------------b3df404c_f41b_42ea_9724_275fd89978ad-1

Three reasons why developers should use DuckDB

=============================================

2: 본 블로그 글은 다양한 LLM(대규모 언어 모델) 추론 엔진인 TensorRT-LLM, vLLM, LMDeploy, MLC-LLM을 비교하고 평가합니다. 각각의 엔진이 LLM 추론 및 서버 성능 최적화를 위해 제공하는 기능과 성능 지표(지연 시간, 처리량 등)를 상세히 분석하여 최적의 추론 결과를 제공합니다. TensorRT-LLM은 NVIDIA GPU를 활용한 고성능 최적화 엔진으로, INT8 모델이 가장 빠른 추론 속도를 보였습니다. vLLM은 최신 스루풋과 CUDA 커널 최적화 기능을 가지고 있으며, LMDeploy는 다중 모델 서비스 전개와 높은 요청 스루풋을 제공합니다. MLC-LLM은 고성능 배포와 추론 엔진을 제공하며, MLC 포맷의 모델 가중치를 변환하여 로드합니다.

키워드: TensorRT-LLM, vLLM, LMDeploy, MLC-LLM

출처: https://medium.com/@zaiinn440/best-llm-inference-engine-tensorrt-vs-vllm-vs-lmdeploy-vs-mlc-llm-e8ff033d7615?source=email-5cbf792c976b-1721149920634-digest.reader--e8ff033d7615----5-98------------------b3df404c_f41b_42ea_9724_275fd89978ad-1

Best LLM Inference Engine? TensorRT vs vLLM vs LMDeploy vs MLC-LLM

=============================================

3: 이 글은 음악 정보 연구(MIR)에서 그래프 신경망(Graph Neural Networks, GNN)을 활용하여 음악 이해 작업을 개선하는 방법에 대해 다룹니다. 특히, 저자는 음악 점수 데이터를 처리하기 위해 설계된 새로운 그래프 컨볼루션 블록인 MusGConv를 소개합니다. MusGConv는 음악의 음높이와 리듬과 같은 지각적 원칙을 활용하여 GNN의 효율성과 성능을 향상시킵니다. MusGConv는 음표 간의 거리(시작 시간, 지속 시간, 음높이) 기반으로 엣지 특징을 계산하고, 절대적 값(음표의 실제 음높이와 시간)과 상대적 특징(음표 간의 음악적 간격)을 결합하여 더욱 의미 있는 음악 데이터 표현을 수행합니다. 이 블록은 다양한 음악 이해 작업(예

키워드: Graph Neural Networks, Music Information Research, MusGConv, 음표 그래프, 메시지 패싱

출처: https://medium.com/@manoskary/perception-inspired-graph-convolution-for-music-understanding-tasks-4d2ba1be48e7?source=email-5cbf792c976b-1721149920634-digest.reader-7f60cf5620c9-4d2ba1be48e7----7-102------------------b3df404c_f41b_42ea_9724_275fd89978ad-1

Perception-Inspired Graph Convolution for Music Understanding Tasks

=============================================

4: 이 글에서는 굉장히 빠른 알고리즘을 사용하여 하나의 DOM 트리를 다른 트리로 변환하는 방법과 이를 통해 DOM을 최소한의 예측 가능한 작업 세트로 자동으로 업데이트하는 방법을 설명합니다. 알고리즘은 DOM 직접 작업 대신 추상 레이어를 사용하며, 주로 Workers나 Node.js에서 실행됩니다. 주요 작업으로는 노드 추가, 삽입, 이동, 제거 등이 있습니다. 이동 작업은 특히 DOM 재생성을 피하면서도 빠른 응답성과 CSS 전환 효과를 제공할 수 있습니다.

키워드: DOM 트리, delta 업데이트, 이동 작업

출처: https://medium.com/@tobiasuhlig/a-blazing-fast-algorithm-to-transform-one-dom-tree-into-another-4ff9c934bc49?source=email-5cbf792c976b-1721149920634-digest.reader-5b301f10ddcd-4ff9c934bc49----9-98------------------b3df404c_f41b_42ea_9724_275fd89978ad-1

A blazing-fast algorithm to transform one DOM tree into another

=============================================

5: 이 글은 Raspberry Pi 5와 Hailo Edge AI 모듈을 사용하여 실시간 웹캠 스트림에서 이미지 탐지를 수행한 후, 탐지된 이미지를 슬랙 채널에 전송하고 메타데이터와 함께 Milvus 벡터 데이터베이스에 저장하는 과정을 설명합니다. 라즈베리파이 AI 키트를 설정하고 Python 라이브러리를 사용하여 객체 감지 프로그램을 수정하고, 탐지된 이미지를 저장 및 전송하는 방법과 관련된 상세한 예시와 코드가 포함되어 있습니다.

키워드: Raspberry Pi 5, Hailo Edge AI, Deep Learning, Milvus, Python

출처: https://medium.com/@tspann/unstructured-data-processing-with-a-raspberry-pi-ai-kit-c959dd7fff47?source=email-5cbf792c976b-1721149920634-digest.reader--c959dd7fff47----10-98------------------b3df404c_f41b_42ea_9724_275fd89978ad-1

Unstructured Data Processing with a Raspberry Pi AI Kit

=============================================

6: 이 글은 FAANG(페이스북, 아마존, 애플, 넷플릭스, 구글 같은 대형 기술 기업)과 스타트업 사이에서 데이터 과학자로서 어떤 회사에 입사할지를 결정하는 데 도움이 되는 여러 요인들을 다루고 있습니다. 저자인 Torsten Walbaum은 Uber와 Meta에서의 경험을 바탕으로 각 회사 유형의 장단점을 설명하고 있으며, 회사의 명성, 동료의 뛰어남, 수입, 리스크, 업무 범위, 학습 기회, 경력 성장 기회 및 스트레스 등을 주요 요인으로 언급합니다. 이를 통해 독자가 자신의 가치와 환경에 맞는 회사를 선택하는 데 도움을 주고자 합니다.

키워드: FAANG, 스타트업, 데이터 과학자

출처: https://medium.com/@twalbaum/should-you-join-faang-or-a-startup-as-a-data-scientist-030e3b8a7080?source=email-5cbf792c976b-1721149920634-digest.reader-7f60cf5620c9-030e3b8a7080----11-98------------------b3df404c_f41b_42ea_9724_275fd89978ad-1

Should You Join FAANG or a Startup as a Data Scientist?

=============================================

7: 이 글에서는 최근 빅 테크 기업들이 인공지능(AI)에 대규모 투자를 하고 있으며, 이는 주로 소수의 거대 기업들이 AI 기술의 독점적 소유권을 확보하기 위한 것이라고 주장하고 있습니다. 저자는 이러한 투자가 결국 소규모 스타트업이나 일반 사용자들에게 불리하게 작용할 것이라고 지적하고 있습니다.

키워드: 인공지능, 벤처 캐피탈, 기술 독점

출처: https://medium.com/@jproco/big-tech-is-spending-a-ton-of-money-on-ai-to-bet-against-you-29a01c747cc0?source=email-5cbf792c976b-1721149920634-digest.reader-7adf33e44ae3-29a01c747cc0----12-99------------------b3df404c_f41b_42ea_9724_275fd89978ad-1

Big Tech Is Spending a Ton Of Money On AI To Bet Against You

=============================================

8: 최근 Apple Worldwide Developers Conference (WWDC)에서 Apple은 Swift 6.0을 공식적으로 출시했습니다. Swift 3 이후로 가장 큰 변화를 이끌어낸 이번 버전은 많은 새로운 기능들과, 이전 버전에서 플래그 뒤에 숨겨졌던 기능들이 기본적으로 활성화되며 개발자들이 쉽게 알아채지 못할 수도 있습니다. 주요 기능으로는 선택적 기능으로 제공되는 컴파일 타임 데이터 레이스 안전성과 마이크로컨트롤러 및 기타 내장 시스템을 위한 Embedded Swift가 포함됩니다. Apple의 Ted Kremenek은 Swift가 C++를 대체할 최고의 프로그래밍 언어라고 말하며, Swift의 안전성, 속도, 사용 편의성, 그리고 내장된 C 및 C++ 상호 운용성을 강조했습니다.

키워드: Swift 6.0, 데이터 레이스 안전성, Embedded Swift

출처: https://medium.com/@dylan_cooper/the-era-of-swift-6-has-arrived-its-the-best-choice-over-c-4963ac8246be?source=email-5cbf792c976b-1721149920634-digest.reader-5b301f10ddcd-4963ac8246be----13-98------------------b3df404c_f41b_42ea_9724_275fd89978ad-1

The Era of Swift 6 Has Arrived! It’s the Best Choice Over C++

=============================================

9: Tenyks 블로그 글은 MLLMs(다중모드 대형 언어 모델)이 컴퓨터 비전 분야를 어떻게 변혁하고 있는지에 대해 설명합니다. MLLM은 GPT-3와 같은 대형 언어 모델의 추론 능력과 시각적 정보 처리를 결합한 모델입니다. 이 글에서는 MLLM의 정의, 주요 사용 사례, 주요 모델 및 최신 연구 동향에 대해 다룹니다. GPT-4o, Claude 3.5 Sonnet 등 여러 최신 모델을 비교 실험하고, 각각의 성능에 대해 분석한 결과를 제시합니다. 특히 Apple의 Ferret 모델이 뛰어난 성능을 보인 점을 강조합니다.

키워드: Multimodal Large Language Model, 컴퓨터 비전, GPT-4o, Apple Ferret, LLaVA

출처: https://medium.com/@tenyks_blogger/multimodal-large-language-models-mllms-transforming-computer-vision-76d3c5dd267f?source=email-5cbf792c976b-1721149920634-digest.reader--76d3c5dd267f----14-98------------------b3df404c_f41b_42ea_9724_275fd89978ad-1

Multimodal Large Language Models (MLLMs) transforming Computer Vision

=============================================

10: 이 글은 서버 인프라를 3-tier 구조로 구성하기 위해 SpringBoot와 MySQL이 함께 있는 EC2 인스턴스를 분리하여 MySQL을 별도의 EC2 인스턴스로 이전하는 과정을 다룹니다. 새로운 EC2 인스턴스를 생성하고, MySQL 서버를 설치 및 설정하며, 기존 EC2의 DB를 덤프한 후 전송합니다. 마지막으로 SpringBoot 프로젝트의 DB 경로를 수정하여 새로 구성된 DB와 연결하고, API 동작을 통해 정상 작동 여부를 확인합니다.

키워드: EC2, MySQL, SpringBoot, 3-tier, RESTful API

출처: https://velog.io/@dradnats1012/DB-%EC%84%9C%EB%B2%84-%EB%B6%84%EB%A6%AC%ED%95%98%EA%B8%B0

DB 서버 분리하기!