Search

On-device training is essential for neural networks (NNs) to continuously adapt to new online data, but can be time-consuming due to the device's limited computing power. To speed up on-device training, existing schemes select trainable NN portion offline or conduct unrecoverable selection at runtime, but the evolution of trainable NN portion is constrained and cannot adapt to the current need for training. Instead, runtime adaptation of on-device training should be fully elastic, i.e., every NN substructure can be freely removed from or added to the trainable NN portion at any time in training. In this paper, we present ElasticTrainer, a new technique that enforces such elasticity to achieve the required training speedup with the minimum NN accuracy loss. Experiment results show that ElasticTrainer achieves up to 3.5x more training speedup in wall-clock time and reduces energy consumption by 2x-3x more compared to the existing schemes, without noticeable accuracy loss., Comment: published at ACM MobiSys 2023. 14 pages

Large Language Models (LLMs) are capable of reasoning over diverse input data modalities through pre-trained encoders. However, the growing diversity of input data modalities prevents incorporating all modalities into LLMs, especially when LLMs are deployed on resource-constrained edge devices for embodied AI applications. Instead, a better option is to adaptively involve only the useful modalities at runtime, depending on the current environmental contexts and task requirements. For such modality adaptation, existing work adopts fixed connections between encoders and the LLM's input layer, leading to high training cost at runtime and ineffective cross-modal interaction. In this paper, we address these limitations by presenting mPnP-LLM, a new technique that allows fully elastic, automated and prompt runtime modality adaptation, by connecting unimodal encoders to a flexible set of last LLM blocks and making such latent connections fully trainable at runtime. Experiments over the nuScenes-QA dataset show that mPnP-LLM can achieve up to 3.7x FLOPs reduction and 30% GPU memory usage reduction, while retaining on-par accuracy with the existing schemes. Under the same compute budget, mPnP-LLM improves the task accuracy by up to 4% compared to the best existing scheme.

In finite element analysis, nonlinear time-history analysis is a realistic and accurate analysis type for dynamic or seismic analysis due to its solutions contain wealthy data and complete response time-history. The most commonly used method, probably the only practical procedure, in nonlinear time-history analysis is the direct time integration method. In general, every direct time integration method could be classified as either an implicit method or an explicit method. Understanding the differences between the two categories in both theoretical and practical aspects is very important for engineers to make the best analysis strategy for a specific dynamic or seismic analysis. In this treatise, the most popular method in each category, i.e., implicit Newmark method and explicit central difference method, will be introduced and used in transient analyses and results comparisons. In total, five cases studies are included in this thesis, including three cases with linear elastic materials and two cases with quasi-brittle masonry material. These five cases are studied to answer the main research questions of this research: What differences can be observed in comparisons of solutions obtained from implicit and explicit methods for linear elastic material in transient analysis and for quasi-brittle material under seismic load? Also, how are the performances of both methods with respect to the stability and accuracy aspects? The results comparisons show that for linear elastic material, both methods could generate accurate solutions with proper time steps. However, the explicit method shows a better representation of the high-frequency response of the structure. For quasi-brittle masonry material, both methods are possible to generate good results. However, the implicit method may have convergence problems during the iterations, which could lead to inaccurate predictions of nonlinear behavior of the structure. The explicit method shows very good predictions for nonlinear

In finite element analysis, nonlinear time-history analysis is a realistic and accurate analysis type for dynamic or seismic analysis due to its solutions contain wealthy data and complete response time-history. The most commonly used method, probably the only practical procedure, in nonlinear time-history analysis is the direct time integration method. In general, every direct time integration method could be classified as either an implicit method or an explicit method. Understanding the differences between the two categories in both theoretical and practical aspects is very important for engineers to make the best analysis strategy for a specific dynamic or seismic analysis. In this treatise, the most popular method in each category, i.e., implicit Newmark method and explicit central difference method, will be introduced and used in transient analyses and results comparisons. In total, five cases studies are included in this thesis, including three cases with linear elastic materials and two cases with quasi-brittle masonry material. These five cases are studied to answer the main research questions of this research: What differences can be observed in comparisons of solutions obtained from implicit and explicit methods for linear elastic material in transient analysis and for quasi-brittle material under seismic load? Also, how are the performances of both methods with respect to the stability and accuracy aspects? The results comparisons show that for linear elastic material, both methods could generate accurate solutions with proper time steps. However, the explicit method shows a better representation of the high-frequency response of the structure. For quasi-brittle masonry material, both methods are possible to generate good results. However, the implicit method may have convergence problems during the iterations, which could lead to inaccurate predictions of nonlinear behavior of the structure. The explicit method shows very good predictions for nonlinear

Boyuan Yang,1 Yi Zuo,1 Qin Zou,1 Limei Li,1 Jidong Li,1 Yi Man,2 Yubao Li1 1Research Center for Nano Biomaterials, Analytical and Testing Center, Sichuan University, Chengdu, People’s Republic of China; 2State Key Laboratory of Oral Disease, West China Hospital of Stomatology, Sichuan University, Chengdu, People’s Republic of China Abstract: We incorporated ultrafine polymer fibers into calcium phosphate cement (CPC) to improve the resorption rate of CPC with fiber degradation. Different weight percentages of electrospun poly(ε-caprolactone) fibers (0%, 3%, and 7%, named as ultrafine fiber-incorporated CPC0 [UFICPC0], UFICPC3, and UFICPC7) were included into preset CPC specimens for in vitro immersion in lipase phosphate-buffered solution and long-term in vivo implantation in the femoral condyle of rabbits. The effect of the ultrafine poly(ε-caprolactone) fibers with a diameter ranging from nanometer to micrometer on CPC degradation was evaluated by measuring the pH of the medium, mass loss, porosity, and physiochemical properties. For the in vivo evaluation, histomorphometrical analysis as well as three-dimensional (3D) reconstruction was applied to assess the osteogenic properties of the CPC composite. After in vitro immersion and in vivo implantation, the total porosity and macroporosity as well as the bone formation and ingrowth increased significantly during time in the fiber-incorporated CPC specimens. After 24 weeks of implantation, the degraded space was occupied by newly formed bone, and the UFICPC3 and UFICPC7 composites showed a ~3.5 times higher fraction of bone volume than that of the pristine CPC (UFICPC0). In vitro and in vivo results proved that the introduction of ultrafine degradable fibers within a CPC matrix can be used to improve macroporosity efficiently and enhance CPC degradation and bone ingrowth largely. Keywords: ultrafine fibers, calcium phosphate cement, macroporosity, degraded space, bone ingr

Boyuan Yang,1 Yi Zuo,1 Qin Zou,1 Limei Li,1 Jidong Li,1 Yi Man,2 Yubao Li1 1Research Center for Nano Biomaterials, Analytical and Testing Center, Sichuan University, Chengdu, People’s Republic of China; 2State Key Laboratory of Oral Disease, West China Hospital of Stomatology, Sichuan University, Chengdu, People’s Republic of China Abstract: We incorporated ultrafine polymer fibers into calcium phosphate cement (CPC) to improve the resorption rate of CPC with fiber degradation. Different weight percentages of electrospun poly(ε-caprolactone) fibers (0%, 3%, and 7%, named as ultrafine fiber-incorporated CPC0 [UFICPC0], UFICPC3, and UFICPC7) were included into preset CPC specimens for in vitro immersion in lipase phosphate-buffered solution and long-term in vivo implantation in the femoral condyle of rabbits. The effect of the ultrafine poly(ε-caprolactone) fibers with a diameter ranging from nanometer to micrometer on CPC degradation was evaluated by measuring the pH of the medium, mass loss, porosity, and physiochemical properties. For the in vivo evaluation, histomorphometrical analysis as well as three-dimensional (3D) reconstruction was applied to assess the osteogenic properties of the CPC composite. After in vitro immersion and in vivo implantation, the total porosity and macroporosity as well as the bone formation and ingrowth increased significantly during time in the fiber-incorporated CPC specimens. After 24 weeks of implantation, the degraded space was occupied by newly formed bone, and the UFICPC3 and UFICPC7 composites showed a ~3.5 times higher fraction of bone volume than that of the pristine CPC (UFICPC0). In vitro and in vivo results proved that the introduction of ultrafine degradable fibers within a CPC matrix can be used to improve macroporosity efficiently and enhance CPC degradation and bone ingrowth largely. Keywords: ultrafine fibers, calcium phosphate cement, macroporosity, degraded space, bone ingr