arXiv subjects

Xinyi Chen

Publications and source records attributed to Xinyi Chen.

111 records · Page 7Linked to original sources

arXiv2019

Organic Thermoelectric Textiles for Harvesting Thermal Energy and Powering Electronics

Wearable thermoelectric devices show promises to generate electricity in a ubiquitous, unintermittent and noiseless way for on-body applications. Three-dimensional thermoelectric textiles (TETs) outperform other types in smart textiles owing to their out-of-plane thermoelectric generation and good structural conformability with fabrics. Yet, there has been lack of efficient strategies in scalable manufacture of TETs for sustainably powering electronics. Here, we fabricate organic spacer fabric shaped TETs by knitting carbon nanotube yarn based segmented thermoelectric yarn in large scale. Combing finite element analysis with experimental evaluation, we elucidate that the fabric structure significantly influences the power generation. The optimally designed TET with good wearability and stability shows high output power density of 51.5 mW/m2 and high specific power of 173.3 uW/(g.K) at delta T= 47.5 K. The promising on-body applications of the TET in directly and continuously powering electronics for healthcare and environmental monitoring is fully demonstrated. This work will broaden the research vision and provide new routines for developing high-performance and large-scale TETs toward practical applications.

physics.app-ph↗

arXiv2019

Calibration, Entropy Rates, and Memory in Language Models

Building accurate language models that capture meaningful long-term dependencies is a core challenge in natural language processing. Towards this end, we present a calibration-based approach to measure long-term discrepancies between a generative sequence model and the true distribution, and use these discrepancies to improve the model. Empirically, we show that state-of-the-art language models, including LSTMs and Transformers, are \emph{miscalibrated}: the entropy rates of their generations drift dramatically upward over time. We then provide provable methods to mitigate this phenomenon. Furthermore, we show how this calibration-based approach can also be used to measure the amount of memory that language models use for prediction.

cs.CL↗

arXiv2019

Extreme Tensoring for Low-Memory Preconditioning

State-of-the-art models are now trained with billions of parameters, reaching hardware limits in terms of memory consumption. This has created a recent demand for memory-efficient optimizers. To this end, we investigate the limits and performance tradeoffs of memory-efficient adaptively preconditioned gradient methods. We propose extreme tensoring for high-dimensional stochastic optimization, showing that an optimizer needs very little memory to benefit from adaptive preconditioning. Our technique applies to arbitrary models (not necessarily with tensor-shaped parameters), and is accompanied by regret and convergence guarantees, which shed light on the tradeoffs between preconditioner quality and expressivity. On a large-scale NLP model, we reduce the optimizer memory overhead by three orders of magnitude, without degrading performance.

cs.LG↗