
고대식당도굴기-챕터 2이문설농탕 In a nutshell김두한 형님의 맛집이자, 김두한이 아르바이트를 했던 곳. 한국에서 제일 오래된 식당 (이라고 알려져 있다).깍두기가 매우 맛있다. 나 깍두기 잘 안 먹는데. (그래서 신빙성 없을지도)뭔가 통통한 밥알.무난히 맛있는 설렁탕.입구"고대 종로"의 느낌이 잘 훈제되어 있는 간판. 물론, 이문설농탕 (123세) 은 2011년에 지금 집으로 이사했다. 지금은 무너졌다고 하는 100년 된 한옥집이 궁금해지는 시점이다. 더 들어가면, 입구. 충분히 낡지 않아서 아쉽달까. 언제부터 낡은 것들이 선호되기 시작했을까? 모두가 여유로워지고부터 일까 아니면 지금이 싫어서일까? 어느 쪽 일지 궁금하다. 내부와 음식 "차림표"의 느낌은 압권이다. 아마 이전 가게에서 그대로 가..

밴건디 in a nutshell서래마을 초입의 -로칼-스테이크 집: 근처사는 가족단위 손님이 대부분인 편임.합리적인 가격: 울프강을 생각하다 보니, 이게 혜자라고 생각이 들어 버린다."맛"집: 로메인 샐러드가 매우 맛있다. 한입에 로메인 한장 가득 넣으면 죽것다. 티본스테이크의 "안심"이 충격이다. 맛있음.Rating: 4.5 / 5.0입구 발렛 차량을 왜 창문 옆에다 붙여두는걸까? 밖이 안보임...음식, 스테이크 이전 27만 5천원짜리 2인 세트같은걸 시켰다. 스테이크 이전에 샐러드 + 파스타 하나를 선택할 수 있었음. 로메인 시저 샐러드 + 통베이컨이 일종의 "시그니처" 인듯 한데, 로메인에 김장하듯 시저 샐러드 드레싱을 구석구석 묻히고, 치즈 그라인딩 해서 올리는듯. 로메인이 드레싱을 먹으면서 적..

ICLR2025논문의 내용과 상관없이, 제가 이야기해보고 싶은 부분은 이탤릭체로 남깁니다.texts with Italic represent parts that are not the part of the original paper, but rather my personal thoughts or additionsLimitation of the LoRALoRA and most of its variants do not perform well when applied to complex tasks—such as commonsense reasoning.This degradation might be inevitable since the additional trainable parameters are “lower-ran..
Creational design patterns and their motivationCreational design pattern: Instance 생성과 관련된 패턴. 어떤 Instance를 효율적으로 생성하고 관리할 수 있게 한다. Creational pattern은 위 처럼 instance 생성이라는 목적으로 만들어진 것들이다. 그렇다면, 모델 학습 파이프라인 등을 구성하는데에는 어떠한 예시가 있을 수 있을까? Instance 생성이 효과적이고, 직관적으로 되는 케이스가 어떤 것들이 있을까?HuggingFace의 AutoModel 같은 경우를 생각해 보자. 이는 (End-user의 입장에서) 모델 instance를 생성하는 데에 굉장히 편리한 역할을 한다: AutoModel.from_pretrain..

IntroModern foundation models for image and video generation are often closed-source, with many technical details remaining unavailable to the public—this is especially true for video generation models. (In my view, this closed-source trend is even more pronounced than in the LLM domain.). For example, we know very little, if anything, about the technical specifics behind models like Runway Gen3..

Summary("Lookahead-loss") During encoder training for models like IP-adaptor, they use single-pass generation models such as the Consistency model to generate noisy images, which are then compared with the reference image to compute an additional loss (LCM-lookahead loss).However, the LCM model can end up optimizing the loss regardless of the input (z_rt below), which could break the alignment b..

Link , Personalized text-to-image generation by Meta AIInsightsThey create “architecture tailored for personalization image generation”they design kind of improved IP-adaptor for “any-personality” generation.therefore, the model does not need to be trained for a new subject, unlike LoRA or Dreambooth.meanwhile, other “any-personality generation models” could come with a strong over-fitting behav..

이 포스트는 Academic-reels와 특집 어딘가에 있는... shorts는 아니지만 그렇다고 10분짜리 비디오도 아닌 그런 구성입니다. 정리본 같은 거랄까요되게 Scientific 하게 잘 쓰인 논문입니다. 점수가 매우 높아요. 흥미로운 가설을 설정했고, 가설을 support 하는 좋은 관측들을 했고, 그에 따른 simple-but-effective 한 방법을 제시합니다. 글도 매우 잘 써진 것 같고요.별 5.FAIR, Meta, ICLR 2024 OralObjective and motivation(Our objective) Vision transformer의 마지막 attention layer를 visualize 해보면, 위와 같은 “abnormal patch“ 가 보인다 (semantically..

경험팔이사실 요즘 동네 주변이나 회사 주변 말고 밥집이든 지나가던 잡화상 (매우 예스럽게 말하자면) 이든, 가게들을 돌아다니다 보면 이게 뭔가 해당 가게의 본질---맛있는 음식, 유용하거나 이쁜 물건들---과는 벗어나서, 그냥 놀이나 경험을 판다는 생각을 하게 만드는 경우가 많다. 뭐 다들 한 번쯤은 생각하겠지만,,, 성수동에서의 실제로 살 물건이 있는건 아니지만 신기해서 들어가 보는 팝업스토어라던지, 귀여운 물건들을 보고 와 굉장히 귀엽다라고 말하고서는 사지 않고 뒤돌아서는 소품샵들이라던지 (이거 정말로 궁금한 건데, 소품샵에서의 실구매율은 얼마나 될지 다른 업종들과 비교해서 측정해보고 싶다). 사실은 피시방에서 끓여 먹는 짜계치가 더 맛있지만 평소에 먹어보지 못한 맛과 멋좀 부리려고 가는 다이닝이나-..

NeurIPS 2023, Arxiv Link What they didThey propose "Battle-of-the-Backones"(BoB), benchmarking a diverse suite of (vision) pre-trained models, including classic ImageNet-trained CNN, vision-language models (perhaps CLIP). The target tasks are OOD-generalization, classification, object-detection and Image retrieval.Key observationswith more than 1,500 training runs, they found that CNN pretrained..
- Total
- Today
- Yesterday
- finetuning
- loss-weighting
- Transformer
- 몽중식
- deeplearning4science
- 이문설농탕
- flowmatching
- MachineLearning
- Theme
- ICML
- icml2024
- LLM
- diffusion
- domaingeneralization
- DeepLearning
- vscode
- multiheadattention
- 프렌밀리
- generativemodeling
일 | 월 | 화 | 수 | 목 | 금 | 토 |
---|---|---|---|---|---|---|
1 | 2 | 3 | 4 | 5 | ||
6 | 7 | 8 | 9 | 10 | 11 | 12 |
13 | 14 | 15 | 16 | 17 | 18 | 19 |
20 | 21 | 22 | 23 | 24 | 25 | 26 |
27 | 28 | 29 | 30 |