1 February 2023 in Asian Journal of Pharmaceutical Sciences
Psoriasis is a chronic inflammatory skin disease characterized by erythema, scaling, and skin thickening. Topical drug application is recommended as the first-line treatment. Many formulation strategies have been developed and explored for enhanced topical psoriasis treatment. However, these preparations usually have low viscosity and limited retention on the skin surface, resulting in low drug delivery efficiency and poor patient satisfaction. In this study, we developed the first water-responsive gel (WRG), which has a distinct water-triggered liquid-to-gel phase transition property. Specifically, WRG was kept in a solution state in the absence of water, and the addition of water induced an immediate phase transition and resulted in a high viscosity gel. Curcumin was used as a model drug to investigate the potential of WRG in topical drug delivery against psoriasis. In vitro and in vivo data showed that WRG formulation could not only extend skin retention but also facilitate the drug permeating across the skin. In a mouse model of psoriasis, curcumin loaded WRG (CUR-WRG) effectively ameliorated the symptoms of psoriasis and exerted a potent anti-psoriasis effect by extending drug retention and facilitating drug penetration. Further mechanism study demonstrated that the anti-hyperplasia, anti-inflammation, anti-angiogenesis, anti-oxidation, and immunomodulation properties of curcumin were amplified by enhanced topical drug delivery efficiency. Notably, neglectable local or systemic toxicity was observed for CUR-WRG application. This study suggests that WRG is a promising formulation for topically psoriasis treatment.
8 June 2023 in IEEE Transactions on Parallel and Distributed Systems
Generalized linear models (GLMs) are a widely utilized family of machine learning models in real-world applications. As data size increases, it is essential to perform efficient distributed training for these models. However, existing systems for distributed training have a high cost for communication and often use large batch sizes to balance computation and communication, which negatively affects convergence. Therefore, we argue for an efficient distributed GLM training system that strives to achieve linear scalability, while keeping batch size reasonably low. As a start, we propose P4SGD, a distributed heterogeneous training system that efficiently trains GLMs through model parallelism between distributed FPGAs and through forward-communication-backward pipeline parallelism within an FPGA. Moreover, we propose a light-weight, latency-centric in-switch aggregation protocol to minimize the latency of the AllReduce operation between distributed FPGAs, powered by a programmable switch. As such, to our knowledge, P4SGD is the first solution that achieves almost linear scalability between distributed accelerators through model parallelism. We implement P4SGD on eight Xilinx U280 FPGAs and a Tofino P4 switch. Our experiments show P4SGD converges up to 6.5X faster than the state-of-the-art GPU counterpart.
1 January 2026 in IEEE Transactions on Networking