arXiv subjects

Jacob Phillips

Publications and source records attributed to Jacob Phillips.

3 recordsLinked to original sources

arXiv2024

Pre-Training Multimodal Hallucination Detectors with Corrupted Grounding Data

Multimodal language models can exhibit hallucinations in their outputs, which limits their reliability. The ability to automatically detect these errors is important for mitigating them, but has been less explored and existing efforts do not localize hallucinations, instead framing this as a classification task. In this work, we first pose multimodal hallucination detection as a sequence labeling task where models must localize hallucinated text spans and present a strong baseline model. Given the high cost of human annotations for this task, we propose an approach to improve the sample efficiency of these models by creating corrupted grounding data, which we use for pre-training. Leveraging phrase grounding data, we generate hallucinations to replace grounded spans and create hallucinated text. Experiments show that pre-training on this data improves sample efficiency when fine-tuning, and that the learning signal from the grounding data plays an important role in these improvements.

cs.CL↗

arXiv2024

Out-of-Distribution Detection & Applications With Ablated Learned Temperature Energy

As deep neural networks become adopted in high-stakes domains, it is crucial to identify when inference inputs are Out-of-Distribution (OOD) so that users can be alerted of likely drops in performance and calibration despite high confidence -- ultimately to know when networks' decisions (and their uncertainty in those decisions) should be trusted. In this paper we introduce Ablated Learned Temperature Energy (or "AbeT" for short), an OOD detection method which lowers the False Positive Rate at 95\% True Positive Rate (FPR@95) by $43.43\%$ in classification compared to state of the art without training networks in multiple stages or requiring hyperparameters or test-time backward passes. We additionally provide empirical insights as to why our model learns to distinguish between In-Distribution (ID) and OOD samples while only being explicitly trained on ID samples via exposure to misclassified ID examples at training time. Lastly, we show the efficacy of our method in identifying predicted bounding boxes and pixels corresponding to OOD objects in object detection and semantic segmentation, respectively -- with an AUROC increase of $5.15\%$ in object detection and both a decrease in FPR@95 of $41.48\%$ and an increase in AUPRC of $34.20\%$ in semantic segmentation compared to previous state of the art.

cs.CV↗

arXiv2016

Moving the Needle: What Works Cities and the use of data and evidence

Bloomberg Philanthropies launched What Works Cities (WWC) in 2015 to help cities better leverage data and evidence to drive decision-making and improve residents' lives. Over three years, WWC will work with 100 American cities with populations between 100,000 and 1,000,000 to measure their state of practice and provide targeted technical assistance. This paper uses the data obtained through the WWC discovery process to understand how 67 cities are currently using data to deliver city services. Our analysis confirms that while cities possess a strong desire to use data and evidence, government leaders are constrained in their ability to apply these practices. We find that a city's stated commitment to using data is the strongest predictor of overall performance and that strong practice in almost any one specific technical area of using data to inform decisions is an indicator of strong practices in other areas. The exception is open data; we find larger cities are more adept at adopting open data policies and programs, independent of their performance using data overall. This paper seeks to develop a deeper understanding of the issues underlying these findings and to continue the conversation on how to best support cities' efforts in this work.

cs.CY↗