Third-party cyber evaluations involving OpenAI models
openai.com·14h ago
AI Research
arXiv CS.AI·May 21·5 min read
Needles in the Landscape: Semi-Supervised Pseudolabeling for Archaeological Site Discovery under Label Scarcity
TL;DR
Archaeological predictive modeling faces challenges due to the scarcity of confirmed site locations, leading to a positive-unlabeled (PU) learning problem. Asymmetric dual pseudolabeling (DPL) is introduced as a deep learning method that effectively utilizes sparse positive data to identify potential archaeological sites.
✦ Why It Matters
Engineers and researchers can apply DPL to enhance predictive modeling in various fields with limited labeled data.
Key Takeaways
- Explore DPL for projects with limited labeled data.
- Utilize environmental variables in predictive modeling.
- Consider PU learning challenges in data-scarce domains.
- Apply findings to improve site discovery accuracy.
Full Summary
Archaeological predictive modeling aims to locate undiscovered sites by analyzing known site locations alongside environmental and geospatial variables. However, this process encounters a positive-unlabeled (PU) learning challenge, where confirmed sites are limited and most locations lack labels, complicating the identification of true negatives.
To address this, asymmetric dual pseudolabeling (DPL) was developed as an end-to-end deep learning method that leverages sparse positive data for training. DPL employs a dual approach to generate pseudolabels, improving the model's ability to predict site locations.
Experimental results indicate that DPL outperforms traditional methods, achieving a notable increase in accuracy and site discovery rates. These findings suggest that DPL can be a valuable tool for archaeologists and researchers in site identification, particularly in data-scarce environments.