Search for: All records

Image super-resolution (SR) is widely used on mobile devices to enhance user experience. However, neural networks used for SR are computationally expensive, posing challenges for mobile devices with limited computing power. A viable solution is to use heterogeneous processors on mobile devices, especially the specialized hardware AI accelerators, for SR computations, but the reduced arithmetic precision on AI accelerators can lead to degraded perceptual quality in upscaled images. To address this limitation, in this paper we present SR For Your Eyes (FYE-SR), a novel image SR technique that enhances the perceptual quality of upscaled images when using heterogeneous processors for SR computations. FYESR strategically splits the SR model and dispatches different layers to heterogeneous processors, to meet the time constraint of SR computations while minimizing the impact of AI accelerators on image quality. Experiment results show that FYE-SR outperforms the best baselines, improving perceptual image quality by up to 2x, or reducing SR computing latency by up to 5.6x with on-par image quality. 

Small—but finite—fluid inertia can be leveraged to generate steady flows out of liquid vibrations around an immersed interface. In engineering, external high-frequency drivers $\left({10}^2\text{–}{10}^5\mathrm{Hz}\right)$allow this inertial rectification phenomenon, known as viscous streaming, to be employed in micron-scale devices for precise flow control, particle manipulation, and spatially controlled chemistry. However, beyond artificial settings, streaming has been hypothesized to be accessible by larger-scale biological systems pertaining to lower frequencies. Then millimeter-size organisms that oscillate or pulsate cilia and appendages in the 1 to $10\mathrm{Hz}$range may be able to rectify surrounding flows, for feeding or locomotion, removing the need for external actuators, tethers, or tubing. Motivated by this potential for bio-hybrid robotic applications and biophysical exploration, here we demonstrate an living system able to produce streaming flows endogenously, autonomously, and unassisted. Computationally informed, our biological device generates oscillatory flows through the cyclic contractions of an engineered muscle tissue, shaped in the form of a torus and suspended in fluid within a microparticle image velocimetry setup. Flow patterns consistent with streaming simulations are observed for low-frequency muscle contractions $(2–4\mathrm{Hz})$, either spontaneous or light-induced, illustrating system autonomy and controllability, respectively. Thus, by connecting tissue engineering with hydrodynamics, this work provides experimental evidence of biologically powered streaming in untethered, millimeter-scale living systems, endowing bio-hybrid technology with inertial microfluidic capabilities. It also illustrates the potential of combining bio-hybrid platforms and simulations to advance both biophysical understanding and fluid mechanics. Published by the American Physical Society2025 

Abstract Ischemia-reperfusion injury (IRI) poses significant challenges across various organ systems, including the heart, brain, and kidneys. Exosomes have shown great potentials and applications in mitigating IRI-induced cell and tissue damage through modulating inflammatory responses, enhancing angiogenesis, and promoting tissue repair. Despite these advances, a more systematic understanding of exosomes from different sources and their biotransport is critical for optimizing therapeutic efficacy and accelerating the clinical adoption of exosomes for IRI therapies. Therefore, this review article overviews the administration routes of exosomes from different sources, such as mesenchymal stem cells and other somatic cells, in the context of IRI treatment. Furthermore, this article covers how the delivered exosomes modulate molecular pathways of recipient cells, aiding in the prevention of cell death and the promotions of regeneration in IRI models. In the end, this article discusses the ongoing research efforts and propose future research directions of exosome-based therapies. Graphical Abstract 

Fine-tuning is essential to adapting pre-trained large language models to downstream applications. With the increasing popularity of LLM-enabled applications, fine-tuning has been performed intensively worldwide, incurring a tremendous amount of computing costs that correspond to big carbon footprint and environmental impact. Mitigating such environmental impact directly correlates to reducing the fine-tuning FLOPs. Existing fine-tuning schemes focus on either saving memory or reducing the overhead of computing weight updates, but cannot achieve sufficient FLOPs reduction due to their ignorance of the training cost in backpropagation. To address this limitation, in this paper we present GreenTrainer, a new technique that minimizes the FLOPs of LLM fine-tuning via adaptive backpropagation, which adaptively selects the most appropriate set of LLM tensors for fine-tuning based on their importance and backpropagation cost in training. Experiment results show that GreenTrainer can save up to 64% training FLOPs compared to full fine-tuning, without any noticeable accuracy loss. Compared to the existing schemes such as Prefix Tuning and LoRA, GreenTrainer can achieve up to 4% improvement of model accuracy, with on-par FLOPs reduction.