arXiv subjects

Hoyong Choi

Publications and source records attributed to Hoyong Choi.

3 recordsLinked to original sources

arXiv2024

BWS: Best Window Selection Based on Sample Scores for Data Pruning across Broad Ranges

Data subset selection aims to find a smaller yet informative subset of a large dataset that can approximate the full-dataset training, addressing challenges associated with training neural networks on large-scale datasets. However, existing methods tend to specialize in either high or low selection ratio regimes, lacking a universal approach that consistently achieves competitive performance across a broad range of selection ratios. We introduce a universal and efficient data subset selection method, Best Window Selection (BWS), by proposing a method to choose the best window subset from samples ordered based on their difficulty scores. This approach offers flexibility by allowing the choice of window intervals that span from easy to difficult samples. Furthermore, we provide an efficient mechanism for selecting the best window subset by evaluating its quality using kernel ridge regression. Our experimental results demonstrate the superior performance of BWS compared to other baselines across a broad range of selection ratios over datasets, including CIFAR-10/100 and ImageNet, and the scenarios involving training from random initialization or fine-tuning of pre-trained models.

cs.LG↗

arXiv2023

Data Valuation Without Training of a Model

Many recent works on understanding deep learning try to quantify how much individual data instances influence the optimization and generalization of a model. Such attempts reveal characteristics and importance of individual instances, which may provide useful information in diagnosing and improving deep learning. However, most of the existing works on data valuation require actual training of a model, which often demands high-computational cost. In this paper, we provide a training-free data valuation score, called complexity-gap score, which is a data-centric score to quantify the influence of individual instances in generalization of two-layer overparameterized neural networks. The proposed score can quantify irregularity of the instances and measure how much each data instance contributes in the total movement of the network parameters during training. We theoretically analyze and empirically demonstrate the effectiveness of the complexity-gap score in finding `irregular or mislabeled' data instances, and also provide applications of the score in analyzing datasets and diagnosing training dynamics. Our code is publicly available at https://github.com/JJchy/CG_score

cs.LG↗

arXiv2021

Power of Bonus in Pricing for Crowdsourcing

We consider a simple form of pricing for a crowdsourcing system, where pricing policy is published a priori, and workers then decide their task acceptance. Such a pricing form is widely adopted in practice for its simplicity, e.g., Amazon Mechanical Turk, although additional sophistication to pricing rule can enhance budget efficiency. With the goal of designing efficient and simple pricing rules, we study the impact of the following two design features in pricing policies: (i) personalization tailoring policy worker-by-worker and (ii) bonus payment to qualified task completion. In the Bayesian setting, where the only prior distribution of workers' profiles is available, we first study the Price of Agnosticism (PoA) that quantifies the utility gap between personalized and common pricing policies. We show that PoA is bounded within a constant factor under some mild conditions, and the impact of bonus is essential in common pricing. These analytic results imply that complex personalized pricing can be replaced by simple common pricing once it is equipped with a proper bonus payment. To provide insights on efficient common pricing, we then study the efficient mechanisms of bonus payment for several profile distribution regimes which may exist in practice. We provide primitive experiments on Amazon Mechanical Turk, which support our analytical findings.

cs.GT↗