This paper presents a unified Vision-Language Pre-training (VLP) model. The model is unified in that (1) it can be fine-tuned for either vision-language generation (e.g., image captioning) or understanding (e.g., visual question answering) tasks, and (2) it uses a shared multi-layer transformer network for both encoding and decoding, which differs from many existing methods where the encoder and decoder are implemented using separate models. The unified VLP model is pre-trained on a large amount of image-text pairs using the unsupervised learning objectives of two tasks: bidirectional and sequence-to-sequence (seq2seq) masked vision-language prediction. The two tasks differ solely in what context the prediction conditions on. This is controlled by utilizing specific self-attention masks for the shared transformer network. To the best of our knowledge, VLP is the first reported model that achieves state-of-the-art results on both vision-language generation and understanding tasks, as disparate as image captioning and visual question answering, across three challenging benchmark datasets: COCO Captions, Flickr30k Captions, and VQA 2.0. The code and the pre-trained models are available at https://github.com/LuoweiZhou/VLP.

Quantitative methods and modeling (QMM) covers a broad spectrum of tool sets, of which physiologically based models and quantitative clinical pharmacology are most critical for generic drugs. QMM has been increasingly applied by the US Food and Drug Administration (FDA) to facilitating generic drug development and review, and has played a critical role in the modernization of bioequivalence (BE) assessment, especially for locally acting drug products, complex products of other types, and modified-release solid oral dosage forms. QMM has aided the development of novel BE methods, in vitro-only BE approaches, and risk-based evaluations. The future of QMM is model integrated evidence or virtual BE studies that can potentially provide pivotal information for generic drug approval. In summary, QMM is indispensable in modernizing generic drug development, BE assessment, and regulatory decision makings. Regulatory examples demonstrate how QMM can be used in modernizing generic drug development, addressing challenges in BE assessment, and supporting regulatory decision making.New drug development and approval depends on sufficient in vitro and in vivo evidence to support the regulatory assessment of drug product efficacy and safety. A generic drug is approved on the basis of sufficient demonstration of sameness to the corresponding brand drug. For both new and generic drugs, quantitative methods and modeling (QMM) can accelerate product development and regulatory assessment.For both new and generic drug development and approval, a mathematical model can be thought of as a knowledge management system that integrates all scientific understanding and existing data about a drug product regarding its formulation, in vitro/in vivo release, pharmacokinetics (PK), pharmacodynamics (PD), and clinical responses. The discipline of QMM both builds these models and uses them to aid both regulatory and business decisions. QMM in new drug developmentFor new drug development, as shown in Figure 1, the data axis includes information and datasets collected through the full Research and Development course for the active pharmaceutical ingredient (API), formulation, in vitro release, the targeted in vivo release profile, animal and human drug PK, PD response(s), and clinical responses in terms of both efficacy and safety end points. Models describing the relationships between the datasets are captured in the model axis. The use of these models can accelerate decisions and optimize the collection of data.Model-informed drug development (MIDD) under the Prescription Drug User Fee Amendments of 2017 (PDUFA VI) is an initiative to use these models to decrease development uncertainty, cost, time, attritions, and failure rates. It aims to inform drug development and regulatory decision makings by using population PK, dose/exposure-response relationships, and biological and statistical models derived from preclinical and clinical data sources. An MIDD pilot program was launched with goals to provide an opportunity for drug developers and the FDA t...

No abstract

No abstract