arXiv subjects

Seon Ho Kim

Publications and source records attributed to Seon Ho Kim.

4 recordsLinked to original sources

arXiv2026

What Does CLIP Learn for Regional Geolocalization? Probing Visual Cues and Scene Configuration After Adaptation

Large collections of street-view imagery provide rich visual information about urban environments, but extracting fine-grained geographic information from such data remains challenging. In particular, fine-grained regional geolocalization is challenging because nearby areas often share coarse geographic cues. We study regional geolocalization within a metropolitan area and ask whether pretrained CLIP features are sufficient for regional discrimination, and what visual information supports performance after adaptation. Using 9,085 street-view images from eight Greater Los Angeles regions, we compare zero-shot CLIP, frozen-encoder readouts, partial encoder updating, Low-Rank Adaptation (LoRA), and full fine-tuning. Frozen readouts remain near the 39.03% zero-shot accuracy, whereas encoder adaptation achieves 75.94-82.10%. Full fine-tuning also reduces the mean distance to the predicted region center from 12.30 km to 3.86 km. We probe these gains through semantic cue removal, appearance reduction using edge maps and blur, and scene-configuration disruption using patch scrambling. Adapted models achieve higher edge and blur accuracy and switch 42.92-45.56% of predictions after scrambling, compared with 10.79-14.60% for frozen methods. However, adaptation does not improve the fraction of performance retained after appearance reduction, while vegetation and sky remain influential. A Caltech101 control further shows that scrambling sensitivity is not unique to geolocalization. Overall, encoder adaptation substantially improves nearby-region discrimination and is associated with greater sensitivity to intact scene configuration, without evidence that coarse structure alone becomes sufficient for prediction. These conclusions concern viewpoint variation near known locations rather than geographically disjoint generalization.

cs.AI↗

arXiv2026

Operating Multi-Node Full Fine-Tuning on NVIDIA B300: A Field Report on Telemetry-Based Triage, Negative Results, and Operational Hardening

We report operational experience full-fine-tuning a 32.76B-parameter dense model (Qwen3-32B) on 16 x NVIDIA B300 (two nodes, FSDP / ZeRO-3) -- among the first published field accounts on this accelerator. We claim no new algorithm. The individual mechanisms we use are established practice; our contribution is the integrated field experience and a set of calibrated measurements on new hardware. Concretely we offer four practitioner artifacts. (1) A B300-calibrated power-draw triage table that distinguishes compute / communication / data-starvation / checkpoint-or-deadlock / idle by board wattage (utilization% reads 100% during an NCCL hang). (2) A set of honest negative results that dispel common optimization folklore at this scale: a controlled A/B in which per-step NFS reading matches a pretokenized local cache (~53k tok/s) because the corpus fits in page cache and the job is compute-bound; and a reconstruction of an earlier "throughput collapse" as NFS/CPU contention rather than a storage-medium limit. (3) Calibrated 4/8/16-GPU strong-scaling and GPU-hour numbers on B300 (near-linear, as expected in this regime; we report absolute values as reference data). (4) A worked failure case -- an epoch-end NCCL deadlock from per-rank token-packing imbalance -- together with a 2.7-second pre-run invariant gate and an external watcher that turn multi-hour silent failures into instant rejections. This deadlock and its remedy correspond to PyTorch's documented Join / equalize-to-minimum practice; we position our instantiation against that prior art and report the GPU-hours the failure cost and the gate saves. The transferable takeaway is operational, not algorithmic: for data-dependent data-parallel jobs, watch power rather than utilization, and verify invariants before launch -- a passing smoke test is not evidence of a safe full run.

cs.DC↗

arXiv2017

A Review on the Applications of Crowdsourcing in Human Pathology

The advent of the digital pathology has introduced new avenues of diagnostic medicine. Among them, crowdsourcing has attracted researchers' attention in the recent years, allowing them to engage thousands of untrained individuals in research and diagnosis. While there exist several articles in this regard, prior works have not collectively documented them. We, therefore, aim to review the applications of crowdsourcing in human pathology in a semi-systematic manner. We firstly, introduce a novel method to do a systematic search of the literature. Utilizing this method, we, then, collect hundreds of articles and screen them against a pre-defined set of criteria. Furthermore, we crowdsource part of the screening process, to examine another potential application of crowdsourcing. Finally, we review the selected articles and characterize the prior uses of crowdsourcing in pathology.

cs.HC↗

arXiv2017

On Identifying Disaster-Related Tweets: Matching-based or Learning-based?

Social media such as tweets are emerging as platforms contributing to situational awareness during disasters. Information shared on Twitter by both affected population (e.g., requesting assistance, warning) and those outside the impact zone (e.g., providing assistance) would help first responders, decision makers, and the public to understand the situation first-hand. Effective use of such information requires timely selection and analysis of tweets that are relevant to a particular disaster. Even though abundant tweets are promising as a data source, it is challenging to automatically identify relevant messages since tweet are short and unstructured, resulting to unsatisfactory classification performance of conventional learning-based approaches. Thus, we propose a simple yet effective algorithm to identify relevant messages based on matching keywords and hashtags, and provide a comparison between matching-based and learning-based approaches. To evaluate the two approaches, we put them into a framework specifically proposed for analyzing disaster-related tweets. Analysis results on eleven datasets with various disaster types show that our technique provides relevant tweets of higher quality and more interpretable results of sentiment analysis tasks when compared to learning approach.

cs.IR↗