Branch retinal vein occlusion (BRVO) is a leading cause of visual impairment in working-age individuals, though predicting its occurrence from vascular features alone remains challenging. We developed deep learning model to predict BRVO based on pre-onset, metadata-matched fundus hemisection images. This retrospective cohort study included patients diagnosed with unilateral two Korean tertiary centers (2005–2023), using images 27 BRVO-affected eyes paired 81 unaffected hemisections (27...

Machine reasoning has made great progress in recent years owing to large language models (LLMs). In the clinical domain, however, most NLP-driven projects mainly focus on classification or reading comprehension, and under-explore for disease diagnosis due expensive rationale annotation with clinicians. this work, we present a "reasoning-aware" framework that rationalizes diagnostic process via prompt-based learning time- labor-efficient manner, learns reason over prompt-generated rationales....

With the popularity of social media (e.g., Facebook and Flicker), users can easily share their check-in records photos during trips. In view huge number user historical mobility in media, we aim to discover travel experiences facilitate trip planning. When planning a trip, always have specific preferences regarding Instead restricting limited query options such as locations, activities, or time periods, consider arbitrary text descriptions keywords about personalized requirements. Moreover,...

Generally 2% of shoppers make a purchase on the first visit to an online store while other 98% enjoys only window-shopping. To bring people back and close deal, "retargeting" has been vital advertising strategy that leads "conversion" window-shoppers into buyers. As such retargeting is more effective as focused tool, in this paper, we study problem identifying conversion rate for given product its current customers, which important analytics metric process. Compared existing approaches using...

As 98 percent of shoppers do not make a purchase on the first visit, we study problem predicting whether they would come back for later (i.e., conversion prediction). This is important strategizing “retargeting”, example, by sending coupons customers who are likely to convert. For this goal, following two problems, prediction market and predictability customer. First, aims at identifying rate given product its customer behavior modeling, which an analytics metric retargeting process....

Purpose: Necrotizing viral retinitis is a serious eye infection that requires immediate treatment to prevent permanent vision loss. Uncertain clinical suspicion can result in delayed diagnosis, inappropriate administration of corticosteroids, or repeated intraocular sampling. To quickly and accurately distinguish between noninfectious retinitis, we aimed develop deep learning (DL) models solely using noninvasive blood test data. Methods: This cross-sectional study trained DL common serology...

With the popularity of social media (e.g., Facebook and Flicker), users could easily share their check-in records photos during trips. In view huge amount data in media, we intend to discover travel experiences facilitate trip planning. Prior works have been elaborated on mining ranking existing routes from data. We observe that when planning a trip, may some keywords about preference his/her Moreover, diverse set is needed. To provide routes, claim more features Places Interests (POIs)...

Abstract Central serous chorioretinopathy (CSC), characterized by detachment of the macular retina, can cause permanent vision loss in chronic course. Chronic CSC is generally treated with photodynamic therapy (PDT), which costly and quite invasive, results are unpredictable. In a retrospective case–control study design, we developed two-stage deep learning model to predict 1-year outcome PDT using initial multimodal clinical data. The training dataset included 166 eyes an additional...

In sentence classification tasks, additional contexts, such as the neighboring sentences, may improve accuracy of classifier. However, contexts are domain-dependent and thus cannot be used for another task with an inappropriate domain. contrast, we propose use translated sentences domain-free context that is always available regardless We find naive feature expansion translations gains only marginal improvements decrease performance classifier, due to possible inaccurate producing noisy...

Yu Jin Kim, Beong-woo Kwak, Youngwook Reinald Kim Amplayo, Seung-won Hwang, Jinyoung Yeo. Proceedings of the 2022 Conference North American Chapter Association for Computational Linguistics: Human Language Technologies. 2022.

In this paper, we propose to leverage the unique characteristics of dialogues sharing commonsense knowledge across participants, resolve difficulties in summarizing them. We present SICK, a framework that uses inferences as additional context. Compared previous work solely relies on input dialogue, SICK an external model generate rich set and selects most probable one with similarity-based selection method. Built upon SICK++ utilizes supervision, where task generating is added dialogue...

Kyungjae Lee, Sunghyun Park, Hojae Han, Jinyoung Yeo, Seung-won Hwang, Juho Lee. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and 9th International Joint (EMNLP-IJCNLP). 2019.

We aim to leverage human and machine intelligence together for attention supervision. Specifically, we show that annotation cost can be kept reasonably low, while its quality enhanced by self-supervision. this goal, explore the advantage of counterfactual reasoning, over associative reasoning typically used in Our empirical results machine-augmented supervision is more effective than existing methods requiring a higher cost, text classification tasks, including sentiment analysis news categorization.

Algorithmic reasoning refers to the ability understand complex patterns behind problem and decompose them into a sequence of steps towards solution. Such nature algorithmic makes it challenge for large language models (LLMs), even though they have demonstrated promising performance in other tasks. Within this context, some recent studies use programming languages (e.g., Python) express necessary logic solving given instance/question Program-of-Thought) as inspired by their strict precise...

This paper covers a sales forecasting problem on e-commerce sites. To predict product sales, we need to understand customers' browsing behavior and identify whether it is for purchase purpose or not. For this goal, propose new customer model, B2P, of aggregating predictive features extracted from history. We perform experiments real world site show that predictions by our model are consistently more accurate than those existing state-of-the-art baselines.

This paper introduces a simple yet effective data-centric approach for the task of improving persona-conditioned dialogue agents. Prior model-centric approaches unquestioningly depend on raw crowdsourced benchmark datasets such as Persona-Chat. In contrast, we aim to fix annotation artifacts in benchmarking, which is orthogonally applicable any model. Specifically, augment relevant personas improve dataset/agent, by leveraging primal-dual structure two tasks, predicting responses and based...

Active learning can be defined as iterations of data labeling, model training, and acquisition, until sufficient labels are acquired. A traditional view acquisition is that, through iterations, knowledge from human models implicitly distilled to monotonically increase the accuracy label consistency. Under this assumption, most recently trained a good surrogate for current labeled data, which requested based on uncertainty/diversity. Our contribution debunking myth proposing new objective...

To build open-domain chatbots that are able to use diverse communicative skills, we propose a novel framework BotsTalk, where multiple agents grounded the specific target skills participate in conversation automatically annotate multi-skill dialogues. We further present Blended Skill BotsTalk (BSBT), large-scale dialogue dataset comprising 300K conversations. Through extensive experiments, demonstrate our can be effective for systems which require an understanding of skill blending as well...

Transfer learning has been widely utilized to mitigate the data scarcity problem in field of Alzheimer's disease (AD). Conventional transfer relies on re-using models trained AD-irrelevant tasks such as natural image classification. However, it often leads negative due discrepancy between non-medical source and target medical domains. To address this, we present evidence-empowered for AD diagnosis. Unlike conventional approaches, leverage an AD-relevant auxiliary task, namely morphological...

Seungone Kim, Se June Joo, Yul Jang, Hyungjoo Chae, Jinyoung Yeo. Proceedings of the 17th Conference European Chapter Association for Computational Linguistics: System Demonstrations. 2023.