اشترك بالحزمة الذهبية واحصل على وصول غير محدود شمرا أكاديميا
تسجيل مستخدم جديدTowards Better Accuracy-efficiency Trade-offs: Divide and Co-training
212
0
0.0
(
0
)
اسأل ChatGPT حول البحث
ﻻ يوجد ملخص باللغة العربية
The width of a neural network matters since increasing the width will necessarily increase the model capacity. However, the performance of a network does not improve linearly with the width and soon gets saturated. In this case, we argue that increasing the number of networks (ensemble) can achieve better accuracy-efficiency trade-offs than purely increasing the width. To prove it, one large network is divided into several small ones regarding its parameters and regularization components. Each of these small networks has a fraction of the original ones parameters. We then train these small networks together and make them see various views of the same data to increase their diversity. During this co-training process, networks can also learn from each other. As a result, small networks can achieve better ensemble performance than the large one with few or no extra parameters or FLOPs. Small networks can also achieve faster inference speed than the large one by concurrent running on different devices. We validate our argument with 8 different neural architectures on common benchmarks through extensive experiments. The code is available at url{https://github.com/mzhaoshuai/Divide-and-Co-training}.
قيم البحث
اقرأ أيضاً
A robot can invoke heterogeneous computation resources such as CPUs, cloud GPU servers, or even human computation for achieving a high-level goal. The problem of invoking an appropriate computation model so that it will successfully complete a task w
hile keeping its compute and energy costs within a budget is called a model selection problem. In this paper, we present an optimal solution to the model selection problem with two compute models, the first being fast but less accurate, and the second being slow but more accurate. The main insight behind our solution is that a robot should invoke the slower compute model only when the benefits from the gain in accuracy outweigh the computational costs. We show that such cost-benefit analysis can be performed by leveraging the statistical correlation between the accuracy of fast and slow compute models. We demonstrate the broad applicability of our approach to diverse problems such as perception using neural networks and safe navigation of a simulated Mars rover.
The goal of this paper is to serve as a guide for selecting a detection architecture that achieves the right speed/memory/accuracy balance for a given application and platform. To this end, we investigate various ways to trade accuracy for speed and
memory usage in modern convolutional object detection systems. A number of successful systems have been proposed in recent years, but apples-to-apples comparisons are difficult due to different base feature extractors (e.g., VGG, Residual Networks), different default image resolutions, as well as different hardware and software platforms. We present a unified implementation of the Faster R-CNN [Ren et al., 2015], R-FCN [Dai et al., 2016] and SSD [Liu et al., 2015] systems, which we view as meta-architectures and trace out the speed/accuracy trade-off curve created by using alternative feature extractors and varying other critical parameters such as image size within each of these meta-architectures. On one extreme end of this spectrum where speed and memory are critical, we present a detector that achieves real time speeds and can be deployed on a mobile device. On the opposite end in which accuracy is critical, we present a detector that achieves state-of-the-art performance measured on the COCO detection task.
This paper is a study of performance-efficiency trade-offs in pre-trained models for automatic speech recognition (ASR). We focus on wav2vec 2.0, and formalize several architecture designs that influence both the model performance and its efficiency.
Putting together all our observations, we introduce SEW (Squeezed and Efficient Wav2vec), a pre-trained model architecture with significant improvements along both performance and efficiency dimensions across a variety of training setups. For example, under the 100h-960h semi-supervised setup on LibriSpeech, SEW achieves a 1.9x inference speedup compared to wav2vec 2.0, with a 13.5% relative reduction in word error rate. With a similar inference time, SEW reduces word error rate by 25-50% across different model sizes.
Trade-offs between accuracy and efficiency are found in multiple non-computing domains, such as law and public health, which have developed rules and heuristics to guide how to balance the two in conditions of uncertainty. While accuracy-efficiency t
rade-offs are also commonly acknowledged in some areas of computer science, their policy implications remain poorly examined. Drawing on risk assessment practices in the US, we argue that, since examining accuracy-efficiency trade-offs has been useful for guiding governance in other domains, explicitly framing such trade-offs in computing is similarly useful for the governance of computer systems. Our discussion focuses on real-time distributed ML systems; understanding the policy implications in this area is particularly urgent because such systems, which include autonomous vehicles, tend to be high-stakes and safety-critical. We describe how the trade-off takes shape for these systems, highlight gaps between existing US risk assessment standards and what these systems require in order to be properly assessed, and make specific calls to action to facilitate accountability when hypothetical risks become realized as accidents in the real world. We close by discussing how such accountability mechanisms encourage more just, transparent governance aligned with public values.
Since the global spread of Covid-19 began to overwhelm the attempts of governments to conduct manual contact-tracing, there has been much interest in using the power of mobile phones to automate the contact-tracing process through the development of
exposure notification applications. The rough idea is simple: use Bluetooth or other data-exchange technologies to record contacts between users, enable users to report positive diagnoses, and alert users who have been exposed to sick users. Of course, there are many privacy concerns associated with this idea. Much of the work in this area has been concerned with designing mechanisms for tracing contacts and alerting users that do not leak additional information about users beyond the existence of exposure events. However, although designing practical protocols is of crucial importance, it is essential to realize that notifying users about exposure events may itself leak confidential information (e.g. that a particular contact has been diagnosed). Luckily, while digital contact tracing is a relatively new task, the generic problem of privacy and data disclosure has been studied for decades. Indeed, the framework of differential privacy further permits provable query privacy by adding random noise. In this article, we translate two results from statistical privacy and social recommendation algorithms to exposure notification. We thus prove some naive bounds on the degree to which accuracy must be sacrificed if exposure notification frameworks are to be made more private through the injection of noise.
سجل دخول لتتمكن من نشر تعليقات
التعليقات
جاري جلب التعليقات
سجل دخول لتتمكن من متابعة معايير البحث التي قمت باختيارها
