[Paper Review] How Contextual are Contextualized Word Representations? Comparing the Geometry of BERT, ELMo, and GPT-2 Embeddings (Contextual Embedding)

1. Anisotropy > Contextualized representations are anisotropic in all non-input layers. If word representations from a particular layer were isotropic

2025๋…„ 1์›” 2์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
ยท

[Paper Review] Low-Rank Adaptation of Large Language Models (LoRA)

1. Overall Summary of the whole content LoRA๋Š” ์‚ฌ์ „ํ•™์Šต๋œ ๋ชจ๋ธ์˜ ๊ฐ€์ค‘์น˜๋ฅผ ๊ณ ์ •ํ•˜๊ณ , ๊ฐ ํŠธ๋žœ์Šคํฌ๋จธ ๋ ˆ์ด์–ด์— ํ•™์Šต ๊ฐ€๋Šฅํ•œ low-rank ํ–‰๋ ฌ์„ ์ถ”๊ฐ€ํ•˜์—ฌ ๋‹ค์šด์ŠคํŠธ๋ฆผ ์ž‘์—…์— ์ ์‘ํ•˜๋„๋ก ํ•œ๋‹ค. ๋…ผ๋ฌธ์—์„œ๋Š” LLM์„ ํšจ์œจ์ ์œผ๋กœ fine-tun

2024๋…„ 12์›” 27์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
ยท
post-thumbnail

Naive Bayes - Count vectorization, TF-IDF

Using raw string text for Machine Learning models = "Natural Language Processing" : Supervised learning text tasks 1. Bayes' Theorem Naive Bayes ->

2024๋…„ 12์›” 20์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
ยท
post-thumbnail

Web-Crawling

1. Crawling > ์›นํŽ˜์ด์ง€๋ฅผ ๊ทธ๋Œ€๋กœ ๊ฐ€์ ธ์™€์„œ ๊ทธ ์•ˆ์—์„œ ๋ฐ์ดํ„ฐ๋ฅผ ์ถ”์ถœํ•˜๋Š” ๊ณผ์ • ํฌ๋กค๋Ÿฌ: ํฌ๋กค๋ง ๊ณผ์ •์„ ์ˆ˜ํ–‰ํ•˜๋Š” ์†Œํ”„ํŠธ์›จ์–ด -> data mining, AI ์–ธ์–ด ๋ชจ๋ธ, ๋น…๋ฐ์ดํ„ฐ ๋ถ„์„ ๋“ฑ ๋‹ค์–‘ํ•œ IT ์˜์—ญ์—์„œ ํ•„์ˆ˜์ ์ธ ์—ญํ•  ์ˆ˜ํ–‰ ์ฃผ๋œ ๋ชฉ์ : ์ •๋ณด ์ˆ˜์ง‘ ๋ฐ ๋ถ„๋ฅ˜

2024๋…„ 12์›” 11์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
ยท
post-thumbnail

[Paper Review] Training language models to follow instructions with human feedback (InstructGPT)

1. Overall Summary of the whole content ๊ธฐ์กด GPT-3์™€ ๊ฐ™์€ LLM์€ ํ…์ŠคํŠธ ์ƒ์„ฑ ๋Šฅ๋ ฅ์ด ๋›ฐ์–ด๋‚˜์ง€๋งŒ, ์‚ฌ์šฉ์ž์˜ ๋ช…๋ น์„ ์™„๋ฒฝํ•˜๊ฒŒ ์ดํ•ดํ•˜๊ณ  ๋”ฐ๋ฅด๋Š” ๋Šฅ๋ ฅ์—๋Š” ์—ฌ์ „ํžˆ ํ•œ๊ณ„๊ฐ€ ์žˆ์œผ๋ฉฐ, ์‚ฌ์šฉ์ž์˜ ์˜๋„์™€ ๋ฏธ๋ฌ˜ํ•˜๊ฒŒ ์–ด๊ธ‹๋‚˜๋Š” ๊ฒฐ๊ณผ๋ฅผ ์ƒ์„ฑํ•˜๋Š” ๊ฒฝ์šฐ๊ฐ€ ๋งŽ

2024๋…„ 11์›” 28์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
ยท
post-thumbnail

[Paper Review] Improving Language Understanding by Generative Pre-Training (GPT-1)

1. Overall Summary of the whole content ์ด ํŽ˜์ดํผ๋Š” Generative Pre-Training(GPT) ๋ฐฉ์‹์œผ๋กœ ์ž์—ฐ์–ด์ฒ˜๋ฆฌ ์„ฑ๋Šฅ์„ ๊ฐœ์„ ํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ์ œ์•ˆํ•œ๋‹ค. (๊ทธ ๊ฒฐ๊ณผ, ์ด์ „์˜ ์ง€๋„ํ•™์Šต ๊ธฐ๋ฐ˜ ๋ชจ๋ธ๋ณด๋‹ค ๋” ์ ์€ label data๋กœ ๋†’์€

2024๋…„ 11์›” 21์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
ยท
post-thumbnail

[Paper Review] Sequence to Sequence Learning with Neural Networks

[Reviewed Paper] https://archive.org/details/arxiv-1409.3215

2024๋…„ 11์›” 9์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
ยท
post-thumbnail

Evaluation steps during the implementation of an ECG classification paper

๋…ผ๋ฌธ ๊ตฌํ˜„ ํ”„๋กœ์ ํŠธ์˜ ์ฝ”๋“œ ์ž‘์„ฑ ๋‹จ๊ณ„๋“ค ์ค‘ ๋Œ€๋žต์ ์ธ ์ „์ฒ˜๋ฆฌ ๊ณผ์ •๊ณผ ๋งˆ์ง€๋ง‰์— ์—ฌ๋Ÿฌ ์ง€ํ‘œ๋“ค์„ ๊ณ„์‚ฐํ•˜์—ฌ ํ‰๊ฐ€๋ฅผ ํ•˜๋Š” ๊ณผ์ •์„ ์ค‘์ ์ ์œผ๋กœ ๋งก๊ฒŒ ๋˜์–ด์„œ, ํ‰๊ฐ€์ง€ํ‘œ ๊ด€๋ จ๋œ ๋‚ด์šฉ์— ๋Œ€ํ•ด ๊ณต๋ถ€ํ•˜๋ฉฐ ์˜ˆ์ „์— ํ•™์Šตํ–ˆ๋˜ ๊ธฐ์–ต์ด ์žˆ๋Š” ๋‚ด์šฉ๋“ค์„ ์ƒ๊ธฐ์‹œ์ผœ๋ดค๋‹ค.ํ‰๊ฐ€์ง€ํ‘œ๋Š” ๋ถ„๋ฅ˜(classificati

2024๋…„ 6์›” 20์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
ยท
post-thumbnail

Developing a Deep Learning Model Trainer Using PyTorch

๋”ฅ๋Ÿฌ๋‹ ๋ชจ๋ธ์„ ํ™œ์šฉํ•œ ๋ฌธ์ œ ํ•ด๊ฒฐ์„ ์œ„ํ•ด์„œ๋Š” ํ•ด๋‹นํ•˜๋Š” ๊ธฐ๋Šฅ์„ ์ˆ˜ํ–‰ํ•  ์ˆ˜ ์žˆ๋„๋ก ๊ตฌ์„ฑ๋œ ํ•™์Šต๊ธฐ๊ฐ€ ํ•„์š”ํ•˜๋‹ค. ๋”ฐ๋ผ์„œ ์ง€๋„ ํ•™์Šต์„ ์œ„ํ•œ ํ•™์Šต๊ธฐ๊ฐ€ ๋™์ž‘ํ•˜๊ธฐ ์œ„ํ•ด ํ•„์š”ํ•œ ๊ตฌ์„ฑ ์š”์†Œ๋“ค์„ ์‚ดํŽด๋ณด๊ณ , ํ”„๋กœ์ ํŠธ ๋•Œ ์‚ฌ์šฉํ•  PyTorch ํ”„๋ ˆ์ž„์›Œํฌ๋ฅผ ํ™œ์šฉํ•œ ๊ฐ„๋‹จํ•œ ์ฝ”๋“œ ์ž‘์„ฑ์— ๋Œ€ํ•ด ์•Œ์•„๋ดค

2024๋…„ 6์›” 7์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
ยท
post-thumbnail

Data Preprocessing : One-hot-encoding

1. Intro ํ”„๋กœ์ ํŠธ์—์„œ ๊ตฌํ˜„ํ•˜๊ธฐ๋กœ ํ•œ ๋…ผ๋ฌธ(https://www.sciencedirect.com/science/article/pii/S2666521222000333?via%3Dihub#abs0015)์€ ๋ฐ์ดํ„ฐ(MIT-BIH, PTB-DB) ๋กœ๋“œ ํ›„ ์ „์ฒ˜๋ฆฌ๊ฐ€ ์ง„ํ–‰๋œ

2024๋…„ 5์›” 30์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
ยท
post-thumbnail

PyTorch Fundamentals - In connection with Tensor

1. Introduction Pytorch๋Š”... > GPU์™€ CPU๋ฅผ ์‚ฌ์šฉํ•˜๋Š” ๋”ฅ ๋Ÿฌ๋‹์— ์ตœ์ ํ™”๋œ ํ…์„œ ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ -> ์ตœ์‹  ์—ฐ๊ตฌ ํ™˜๊ฒฝ์—์„œ ์ง€๋ฐฐ์ , ์ปค๋ฎค๋‹ˆํ‹ฐ์—์„œ ๋„๋ฆฌ ์ฑ„ํƒ๋˜๊ณ  ๋Œ€๋ถ€๋ถ„์˜ ์ถœํŒ๋ฌผ/์‚ฌ์šฉ ๊ฐ€๋Šฅํ•œ ๋ชจ๋ธ์—์„œ ์‚ฌ์šฉ ** 2. Tensor (1) Tensor๋ž€

2024๋…„ 5์›” 24์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
ยท
post-thumbnail

[Paper Review] Classifying ECG abnormalities using 1D Convolutional Neural Networks(1D-CNN) with Leaky-ReLU function

1. Summary of this paper The steps of preprocessing(followed by QRS complex detection, data augmentation, data subdivision) and modeling have a signif

2024๋…„ 5์›” 14์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
ยท
post-thumbnail

[Paper Review] ECG Arrhythmia Classification using RNN

1\. Summary of this paperThe primary aim of this paper is to enable automatic seperation applying RNN(Recurrent Neural Networks) to classify the norma

2024๋…„ 5์›” 9์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
ยท