Source author record

Mengze Zhu

Mengze Zhu appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

cond-mat.mtrl-sci Computer Vision cs.CY Machine Learning

Catalog footprint

What is connected

4works

4topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2021arXiv

DeepWear: Adaptive Local Offloading for On-Wearable Deep Learning

Due to their on-body and ubiquitous nature, wearables can generate a wide range of unique sensor data creating countless opportunities for deep learning tasks. We propose DeepWear, a deep learning (DL) framework for wearable devices to improve the performance and reduce the energy footprint. DeepWear strategically offloads DL tasks from a wearable device to its paired handheld device through local network. Compared to the remote-cloud-based offloading, DeepWear requires no Internet connectivity, consumes less energy, and is robust to privacy breach. DeepWear provides various novel techniques such as context-aware offloading, strategic model partition, and pipelining support to efficiently utilize the processing capacity from nearby paired handhelds. Deployed as a user-space library, DeepWear offers developer-friendly APIs that are as simple as those in traditional DL libraries such as TensorFlow. We have implemented DeepWear on the Android OS and evaluated it on COTS smartphones and smartwatches with real DL models. DeepWear brings up to 5.08X and 23.0X execution speedup, as well as 53.5% and 85.5% energy saving compared to wearable-only and handheld-only strategies, respectively.

preprint2020arXiv

DeepCache: Principled Cache for Mobile Deep Vision

We present DeepCache, a principled cache design for deep learning inference in continuous mobile vision. DeepCache benefits model execution efficiency by exploiting temporal locality in input video streams. It addresses a key challenge raised by mobile vision: the cache must operate under video scene variation, while trading off among cacheability, overhead, and loss in model accuracy. At the input of a model, DeepCache discovers video temporal locality by exploiting the video's internal structure, for which it borrows proven heuristics from video compression; into the model, DeepCache propagates regions of reusable results by exploiting the model's internal structure. Notably, DeepCache eschews applying video heuristics to model internals which are not pixels but high-dimensional, difficult-to-interpret data. Our implementation of DeepCache works with unmodified deep learning models, requires zero developer's manual effort, and is therefore immediately deployable on off-the-shelf mobile devices. Our experiments show that DeepCache saves inference execution time by 18% on average and up to 47%. DeepCache reduces system energy consumption by 20% on average.

preprint2015arXiv

Lattice correlation of Hubbard excitons in a Mott insulator Sr2IrO4 and reconstruction of their hopping dynamics via time-dependent coherence analysis of the Bragg diffraction

In correlated oxides the coupling of quasiparticles to other degrees of freedom such as spin and lattice plays critical roles in the emergence of symmetry-breaking quantum ordered states such as high temperature superconductivity. We report a strong lattice coupling of photon induced Hubbard excitonic quasiparticles in spin-orbital coupling Mott insulator Sr2IrO4. Combining time-resolved optical spectroscopy techniques, we further reconstructed spatiotemporal map of the diffusion of quasiparticles via time-dependent coherence analysis of the x-ray Bragg diffraction peak. Due to the unique electronic configuration of the exciton, the strong lattice correlation is unexpected but extends the similarity between Sr2IrO4 and cuprates under highly non-equilibrium conditions. The coherence analysis method we developed may have important implications for characterizing the structure and carrier dynamics in a wider group of oxide heterostructures.

preprint2014arXiv

Excess-Hole Induced High Temperature Polarized State and its Correlation with the Multiferroicity in Single Crystalline DyMnO3

Controlling the ferroelectricity and magnetism in multiferroic materials has been an important research topic. We report the formation of a highly polarized state in multiferroic DyMnO3 single crystals which develops well above the magnetic transition temperatures, and we attribute it the thermally stimulated depolarization current effect of excess holes forming Mn4+ ions in the system. We also show that this high temperature polarized state intimately correlates with the lower temperature ferroelectric state that is induced by the incommensurate spiral magnetic order of Mn spins. This study demonstrate an efficient approach to tune the multiferroicity in the manganite system.