Publications

The recent emergence of Large Language Models based on the Transformer architecture has enabled dramatic advancements in the field of Natural Language Processing. However, these models have long inference latency, which limits their deployment and makes them prohibitively expensive for various real-time applications. The inference latency is further exacerbated by autoregressive generative tasks, as models need to run iteratively to generate tokens sequentially without leveraging token-level parallelization. To address this, we propose Big Little Decoder (BiLD), a framework that can improve inference efficiency and latency for a wide range of text generation applications. The BiLD framework contains two models with different sizes that collaboratively generate text. The small model runs autoregressively to generate text with a low inference cost, and the large model is only invoked occasionally to refine the small model's inaccurate predictions in a non-autoregressive manner. To coordinate the small and large models, BiLD introduces two simple yet effective policies: (1) the fallback policy that determines when to hand control over to the large model; and (2) the rollback policy that determines when the large model needs to correct the small model's inaccurate predictions. To evaluate our framework across different tasks and models, we apply BiLD to various text generation scenarios encompassing machine translation on IWSLT 2017 De-En and WMT 2014 De-En, and summarization on XSUM and CNN/DailyMail. On an NVIDIA T4 GPU, our framework achieves a speedup of up to 2.12x speedup with minimal generation quality degradation. Furthermore, our framework is fully plug-and-play and can be applied without any modifications in the training process or model architecture. Our code is open-sourced

The fracture of bone is clearly a major health concern, especially for the elderly. Medical therapies to reduce the possibility of bone fracture to date have principally centered on treating the loss in bone mass (bone mineral density) that accompanies aging (i.e., addressing the loss in bone quantity). However, it is now known that there is an additional, perhaps more significant, effect of the degradation in the inherent properties of bone (i.e., a loss in bone quality) with age. To address this issue, we review here the structure and properties of bone, focusing on its strength and fracture resistance from the perspective of the multidimensional hierarchical nature of its structure. We show that bone derives its resistance to fracture from a multitude of deformation and toughening mechanisms at many of these size‐scales, ranging from the nanoscale structure of its protein molecules to its macroscopic physiological state.

We relax the long-held and problematic assumption in shape-from-shading (SFS) that albedo must be uniform or known, and address the problem of "shape and albedo from shading" (SAFS). Using models normally reserved for natural image statistics, we impose "naturalness" priors over the albedo and shape of a scene, which allows us to simultaneously recover the most likely albedo and shape that explain a single image. A simplification of our algorithm solves classic SFS, and our SAFS algorithm can solve the intrinsic image decomposition problem, as it solves a superset of that problem. We present results for SAFS, SFS, and intrinsic image decomposition on real lunar imagery from the Apollo missions, on our own pseudo-synthetic lunar dataset, and on a subset of the MIT Intrinsic Images dataset[15]. Our one unified technique appears to outperform the previous best individual algorithms for all three tasks. Our technique allows a coarse observation of shape (from a laser rangefinder or a stereo algorithm, etc) to be incorporated a priori. We demonstrate that even a small amount of low-frequency information dramatically improves performance, and motivate the usage of shading for high-frequency shape (and albedo) recovery.

A fracture‐mechanics based study has performed to characterize the fracture toughness and rates of cyclic fatiguecrack growth of incipient flaws in prosthetic heart‐valve components made of pyrolytic carbon‐coated graphite. Such data are required to predict the safe structural lifetime of mechanical heart‐valve prostheses using damagetolerant analysis. Unlike previous studies where fatiguecrack propagation data were obtained using through thickness, long cracks (∼2–20 mm long), growing in conventional (e.g., compact‐tension) samples, experiments were performed on physically small cracks (∼100–600 μm long), initiated on the surface of the pyrolytic‐carbon coating to simulate reality. Small‐crack toughness results were found to agree closely with those measured conventionally with long cracks. However, similar to well‐known observations in metal fatigue, it was found that based on the usual computations of the applied (far‐field) driving force in terms of the maximum stress intensity, K max , small fatigue cracks grew at rates that exceeded those of long cracks at the same applied stress intensity, and displayed a negative dependency on K max ; moreover, they grew at applied stress intensities less than the fatigue threshold value, below which long cracks are presumed dormant. To resolve this apparent discrepancy, it is shown that long and small crack results can be normalized, provided growth rates are characterized in terms of the total (near‐tip) stress intensity (incorporating, for example, the effect of residual stress); with this achieved, in principle, either form of data can be used for life prediction of implant devices. Inspection of the long and small crack results reveals extensive scatter inherent in both forms of growth‐rate data for the pyrolytic‐carbon material. © 1994 John Wiley & Sons, Inc.

Abstract The use of switched‐capacitor components for designing non‐linear networks is discussed in this paper. We first introduce a new type of network component, the BESC resistor, and consider its application to the design of non‐linear transfer characteristic and driving point plots. Systematic methods for designing piecewise‐linear BESC resistors using switched capacitors are then given. Finally, experimental results showing the performance of the proposed circuits are included.