Meta tags:
author=Jae Myung Kim;

Headings (most frequently used words):

with, shot, feasibility, in, for, classification, and, training, dataset, generation, few, the, concept, models, visual, multi, label, jae, myung, kim, selected, papers, loft, lora, fused, guidance, does, matter, understanding, impact, of, on, synthetic, data, datadream, guided, improving, intervention, efficacy, via, realignment, bottleneck, language, open, world, compositional, zero, learning, waffling, around, performance, random, words, broad, concepts, exposing, mitigating, spurious, correlations, cross, modal, retrieval, bridging, gap, between, model, explanations, partially, annotated, large, loss, matters, weakly, supervised, keep, calm, improve, feature, attribution,

Text of the page (most frequently used words):
the (23), data (16), and (15), kim (14), with (13), shot (13), training (12), jae (12), myung (12), zeynep (10), akata (10), arxiv (10), #synthetic (10), code (9), models (9), for (7), cordelia (7), schmid (7), learning (7), few (7), zero (6), classification (5), dataset (5), feasibility (5), multi (4), weak (4), alignment (4), that (4), model (4), performance (4), concept (4), real (4), visual (3), attribution (3), its (3), explanation (3), explainability (3), label (3), task (3), samples (3), cvpr (3), trained (3), labels (3), 2023 (3), spurious (3), correlations (3), bias (3), random (3), stephan (3), alaniz (3), eccv (3), 2024 (3), generate (3), guided (3), how (3), can (3), source (2), from (2), color (2), seong (2), joon (2), website (2), feature (2), but (2), graph (2), this (2), iccv (2), youngwook (2), jungwoo (2), lee (2), partially (2), observe (2), memorization (2), then (2), propose (2), high (2), loss (2), matters (2), weakly (2), full (2), partial (2), each (2), different (2), sophia (2), koepke (2), retrieval (2), learn (2), workshop (2), modal (2), karsten (2), roth (2), achieve (2), access (2), leverage (2), open (2), world (2), compositional (2), intervention (2), reach (2), jessica (2), bader (2), which (2), generation (2), 2025 (2), lora (2), questions (2), have (2), previously (2), google (2), better (2), working (2), munich (2), university (2), scientist (2), design, keywords, coding, inspired, leonid, keselman, junsuk, choe, class, activation, mapping, cam, widely, used, reliance, hoc, calibration, steps, outside, limits, interpretability, address, issue, introducing, latent, variable, cue, location, itself, 2021, keep, calm, improve, frame, labeled, setting, noisy, effect, reject, correct, preventing, noise, 2022, large, supervised, jieun, jeong, two, highlights, similar, regions, scaling, boost, scores, make, resemble, bridging, gap, between, explanations, annotated, find, image, text, commonly, memorize, introduce, metric, measures, robustness, those, finetuning, them, controlled, exposing, mitigating, cross, oriol, vinyals, comparable, clip, without, external, using, characters, word, descriptors, waffling, around, words, broad, concepts, llms, determine, state, object, combinations, language, nishad, singhi, relations, realign, assignments, post, cbms, effectively, reduces, number, necessary, interventions, target, improving, efficacy, via, realignment, bottleneck, more, faithfully, represents, distribution, targeted, datadream, yiwen, liu, study, whether, when, classifiers, conduct, experiments, three, attributes, background, texture, cvprw, syndata4cv, fgvc12, best, paper, award, does, matter, understanding, impact, fusing, weights, corresponding, sample, datasets, fidelity, sufficient, diversity, contribute, improvement, bmvc, loft, fused, guidance, selected, papers, addition, these, worked, building, topics, such, xai, uncertainty, linkedin, scholar, email, reliable, foundation, minimum, are, normally, aligned, representations, exploring, generative, leveraged, robust, pertaining, downstream, tasks, broadly, interested, efficient, centric, approaches, moment, following, phd, student, member, programs, advised, helmholtz, closely, inria, completed, degrees, seoul, national, imprs, ellis, tübingen, currently, seeking, research, applied, roles, industry, love, connect, please, feel, free, out,

Text of the page (random words):
jae myung kim jae myung kim i m currently seeking research scientist or applied scientist roles in the industry i d love to connect please feel free to reach out i am a phd student at the university of tübingen and a member of ellis imprs is programs advised by zeynep akata tu munich helmholtz munich and closely working with cordelia schmid inria google previously i completed my b s and m s degrees at the seoul national university i am broadly interested in efficient data centric approaches at the moment i m working on the following questions synthetic data as training data exploring how generative models can be leveraged as a robust data source for pertaining or downstream tasks weak alignment multi modal data are normally weakly aligned how can we have better representations with weak data alignment zero shot and few shot learning how can we better leverage foundation models with access to minimum data in addition to these questions i have previously worked on building reliable models topics such as xai bias and uncertainty email google scholar linkedin selected papers synthetic training data few shot learning loft lora fused training dataset generation with few shot guidance jae myung kim stephan alaniz cordelia schmid zeynep akata bmvc 2025 arxiv code we generate a synthetic dataset guided by few shot real samples by fusing lora weights corresponding to each real sample we achieve to generate datasets with high fidelity and sufficient diversity which contribute to performance improvement synthetic training data does feasibility matter understanding the impact of feasibility on synthetic training data yiwen liu jessica bader jae myung kim cvprw on syndata4cv fgvc12 2025 best paper award arxiv code we study whether feasibility matters when training classifiers with synthetic data we conduct experiments on three different attributes background color and texture synthetic training data few shot learning datadream few shot guided dataset generation jae myung kim jessica bader stephan alaniz cordelia schmid zeynep akata eccv 2024 arxiv code we generate a synthetic dataset guided by few shot real samples which more faithfully represents the real data distribution of the targeted classification task explainability improving intervention efficacy via concept realignment in concept bottleneck models nishad singhi jae myung kim karsten roth zeynep akata eccv 2024 arxiv code we learn concept relations to realign concept assignments post intervention in cbms this effectively reduces the number of necessary interventions to reach a target performance zero shot learning feasibility with language models for open world compositional zero shot learning jae myung kim stephan alaniz cordelia schmid zeynep akata eccv workshop 2024 arxiv we leverage llms to determine the feasibility of state object combinations for open world compositional zero shot learning task zero shot learning waffling around for performance visual classification with random words and broad concepts karsten roth jae myung kim a sophia koepke oriol vinyals cordelia schmid zeynep akata iccv 2023 arxiv code we achieve comparable zero shot clip performance without access to external models by using random characters and random word descriptors bias synthetic training data exposing and mitigating spurious correlations for cross modal retrieval jae myung kim a sophia koepke cordelia schmid zeynep akata cvpr workshop 2023 arxiv we find that image text retrieval models commonly learn to memorize spurious correlations in the training data we introduce a metric that measures a model s robustness to spurious correlations in the training data and de bias those models by finetuning them with the controlled synthetic dataset weak alignment explainability bridging the gap between model explanations in partially annotated multi label classification youngwook kim jae myung kim jieun jeong cordelia schmid zeynep akata jungwoo lee cvpr 2023 arxiv code we observe that the explanation of two models trained with full and partial labels each highlights similar regions but with different scaling we then propose to boost the attribution scores of the model trained with partial labels to make its explanation resemble that of the model trained with full labels weak alignment large loss matters in weakly supervised multi label classification youngwook kim jae myung kim zeynep akata jungwoo lee cvpr 2022 arxiv code we frame the partially labeled setting as a noisy multi label classification task and observe the memorization effect we then propose to reject or correct high loss samples preventing the memorization of noise explainability keep calm and improve visual feature attribution jae myung kim junsuk choe zeynep akata seong joon oh iccv 2021 arxiv code class activation mapping cam is widely used for visual feature attribution but its reliance on ad hoc calibration steps outside the training graph limits its interpretability we address this issue by introducing a latent variable for cue location an explanation by itself in the training graph design and source code from leonid keselman s website keywords color coding inspired from seong joon oh s website