Search for: All records

In Cloud 3D, such as Cloud Gaming and Cloud Virtual Reality (VR), image frames are rendered and compressed (encoded) in the cloud, and sent to the clients for users to view. For low latency and high image quality, fast, high compression rate, and high-quality image compression techniques are preferable. This paper explores computation time reduction techniques for learned image compression to make it more suitable for cloud 3D. More specifically, we employed slim (low-complexity) and application-specific AI models to reduce the computation time without degrading image quality. Our approach is based on two key insights: (1) as the frames generated by a 3D application are highly homogeneous, application-specific compression models can improve the rate-distortion performance over a general model; (2) many computer-generated frames from 3D applications are less complex than natural photos, which makes it feasible to reduce the model complexity to accelerate compression computation. We evaluated our models on six gaming image datasets. The results show that our approach has similar rate-distortion performance as a state-of-the-art learned image compression algorithm, while obtaining about 5x to 9x speedup and reducing the compression time to be less than 1 s (0.74s), bringing learned image compression closer to being viable for cloudmore »3D. Code is available at https://github.com/cloud-graphics-rendering/AppSpecificLIC.« less

Momentum stochastic gradient descent (MSGD) algorithm has been widely applied to many nonconvex optimization problems in machine learning (e.g., training deep neural networks, variational Bayesian inference, etc.). Despite its empirical success, there is still a lack of theoretical understanding of convergence properties of MSGD. To fill this gap, we propose to analyze the algorithmic behavior of MSGD by diffusion approximations for nonconvex optimization problems with strict saddle points and isolated local optima. Our study shows that the momentum helps escape from saddle points but hurts the convergence within the neighborhood of optima (if without the step size annealing or momentum annealing). Our theoretical discovery partially corroborates the empirical success of MSGD in training deep neural networks.

Surfactant protein D (SP-D) is a C-type collectin and plays an important role in innate immunity and homeostasis in the lung. This study studied SP-D role in the nontypeable Haemophilus influenzae (NTHi)-induced otitis media (OM) mouse model. Wild-type C57BL/6 (WT) and SP-D knockout (KO) mice were used in this study. Mice were injected in the middle ear (ME) with 5 μL of NTHi bacterial solution (3.5 × 10⁵ CFU/ear) or with the same volume of sterile saline (control). Mice were sacrificed at 3 time points, days 1, 3, and 7, after treatment. We found SP-D expression in the Eustachian tube (ET) and ME mucosa of WT mice but not in SP-D KO mice. After infection, SP-D KO mice showed more intense inflammatory changes evidenced by the increased mucosal thickness and inflammatory cell infiltration in the ME and ET compared to WT mice (p < 0.05). Increased bacterial colony-forming units and cytokine (IL-6 and IL-1β) levels in the ear washing fluid of infected SP-D KO mice were compared to infected WT mice. Molecular analysis revealed higher levels of NF-κB and NLRP3 activation in infected SP-D KO compared to WT mice (p < 0.05). In vitro studies demonstrated that SP-D significantly inducedmore »NTHi bacterial aggregation and enhanced bacterial phagocytosis by macrophages (p < 0.05). Furthermore, human ME epithelial cells showed a dose-dependent increased expression of NLRP3 and SP-D proteins after LPS treatment. We conclude that SP-D plays a critical role in innate immunity and disease resolution through enhancing host defense and regulating inflammatory NF-κB and NLRP3 activation in experimental OM mice.« less