A5100575316- China's Socioeconomic Reforms and Governance
- Multimodal Machine Learning Applications
- Domain Adaptation and Few-Shot Learning
- Legal principles and applications
- Chinese history and philosophy
- Immigration Law and Human Rights
- Canadian Identity and History
- Crime Patterns and Interventions
- Advanced Image and Video Retrieval Techniques
- Ombudsman and Human Rights
- Vietnamese History and Culture Studies
- Policing Practices and Perceptions
- Species Distribution and Climate Change
- Video Analysis and Summarization
- Digital Storytelling and Education
- Genomics and Phylogenetic Studies
- Criminal Justice and Corrections Analysis
- Image Retrieval and Classification Techniques
- Topic Modeling
- Property Rights and Legal Doctrine
- Economics of Agriculture and Food Markets
- Subtitles and Audiovisual Media
- Digital Humanities and Scholarship
- Torture, Ethics, and Law
- Identification and Quantification in Food
Carnegie Mellon University
2021-2024
Carleton University
1998-2016
Health Canada
2001
Heilongjiang University
1991
The ability to quickly learn a new task with minimal instruction - known as few-shot learning is central aspect of intelligent agents. Classical benchmarks make use samples from single modality, but such may not be sufficient characterize an entire concept class. In contrast, humans cross-modal information concepts efficiently. this work, we demonstrate that one can indeed build better visual dog classifier by reading about dogs and listening them bark. To do so, exploit the fact recent...
Following the recent popularity of Large Language Models (LLMs), several attempts have been made to extend them visual domain. From having a assistant that could guide us through unfamiliar environments generative models produce images using only high-level text description, vision-language model (VLM) applications will significantly impact our relationship with technology. However, there are many challenges need be addressed improve reliability those models. While language is discrete,...
The analysis of the profile and role China's Supreme People's Court needs updating. is actively developing new interpretative formats that concern its relations with sister organizations National Congress. This article contextualizes these within changing institutional dynamics. China does not have a separation powers; however, Chinese system justice own functions. playing pivotal from functions, but extent quality independence other are open to question. In context deepening legal reform,...
This article examines the CCP's “ falun gong problem” with reference to PRC law and policy on “heretical cults,” paying particular attention implications of this problem for ongoing struggle establish human rights under rule law. Official commentary contends that not only committed criminal acts but also wilfully sought undermine itself. Human critics agencies, such as US Commission International Religious Freedom, have, other hand, attacked a “repressive legal framework” threatens rights....
Trained on web-scale image-text pairs, Vision-Language Models (VLMs) such as CLIP can recognize images of common objects in a zero-shot fashion. However, it is underexplored how to use for recognition highly specialized concepts, e.g., species birds, plants, and animals, which their scientific names are written Latin or Greek. Indeed, performs poorly with prompts that names, "a photo Lepus Timidus" (which name Latin). This because these usually not included CLIP's training set. To improve...
The toxicity of pesticides on human reproduction is largely unknown--particularly how mixtures pesticide products might affect fetal toxicity. Ontario Farm Family Health Study collected data by questionnaire the identity and timing use farm, lifestyle factors, a complete reproductive history from farm operator eligible couples living farm. A total 2,110 women provided information 3,936 pregnancies, including 395 spontaneous abortions. To explore critical windows exposure target sites for...
Vision-language models (VLMs) excel in zero-shot recognition but their performance varies greatly across different visual concepts. For example, although CLIP achieves impressive accuracy on ImageNet (60-80%), its drops below 10% for more than ten concepts like night snake, presumably due to limited presence the pretraining data. However, measuring frequency of VLMs' large-scale datasets is challenging. We address this by using large language (LLMs) count number texts that contain synonyms...
Vision-language models (VLMs) are impactful in part because they can be applied to a variety of visual understanding tasks zero-shot fashion, without any fine-tuning. We study $\textit{generative VLMs}$ that trained for next-word generation given an image. explore their performance on the illustrative task image-text retrieval across 8 popular vision-language benchmarks. Our first observation is repurposed discriminative (such as retrieval) by simply computing match score generating...
This article surveys the Chinese response to SARS in law and politics. Over course of spread party-state qualified legal reform strategy that was designed provide new human rights protection curtail state's arbitrary resort policy regulation without benefit law. immediate revealed underlying problems rule-of-law making, but experience later informed creation improved on infectious disease reiterated original assumptions within a newly developing approach public management health crises.
Generative Large Multimodal Models (LMMs) like LLaVA and Qwen-VL excel at a wide variety of vision-language (VL) tasks such as image captioning or visual question answering. Despite strong performance, LMMs are not directly suited for foundational discriminative (i.e., requiring discrete label predictions) classification multiple-choice VQA. One key challenge in utilizing is the extraction useful features from generative models. To overcome this issue, we propose an approach finding model's...
Despite significant progress in generative AI, comprehensive evaluation remains challenging because of the lack effective metrics and standardized benchmarks. For instance, widely-used CLIPScore measures alignment between a (generated) image text prompt, but it fails to produce reliable scores for complex prompts involving compositions objects, attributes, relations. One reason is that encoders CLIP can notoriously act as "bag words", conflating such "the horse eating grass" with grass...
While text-to-visual models now produce photo-realistic images and videos, they struggle with compositional text prompts involving attributes, relationships, higher-order reasoning such as logic comparison. In this work, we conduct an extensive human study on GenAI-Bench to evaluate the performance of leading image video generation in various aspects generation. We also compare automated evaluation metrics against our collected ratings find that VQAScore -- a metric measuring likelihood VQA...
Vision-language models (VLMs) have made significant progress in recent visual-question-answering (VQA) benchmarks that evaluate complex visio-linguistic reasoning. However, are these truly effective? In this work, we show VLMs still struggle with natural images and questions humans can easily answer, which term adversarial samples. We also find it surprisingly easy to generate VQA samples from image-text corpora using off-the-shelf like CLIP ChatGPT. propose a semi-automated approach collect...
This article discusses in detail the content and context of China’s recent sentencing reform its social, political, criminal justice implications, as well limitations. The focus reforms over past 37 years has been predominantly on trial process; process largely neglected. Revelations widespread inconsistency led Supreme People’s Court (SPC) to initiate 2005. intent was promote transparency process, ensure consistency dispositions, guard against inappropriate judicial leniency severity via...
* This research was made possible by a grant from SSHRCC for an historical study of the prairie provincial police institutions (#410-91-1395). We thank A. R. Gillis, Tullio Caputo, Leslie Miller, as well anonymous reviewer journal helpful comments on earlier draft. Abstract: In late 19th century in North America and Britain prosecuted public order offenses way regulating moral criminal conduct lower classes. Between 1896 1940, law similarly employed to assimilate `dangerous foreigners'...