اشترك بالحزمة الذهبية واحصل على وصول غير محدود شمرا أكاديميا
تسجيل مستخدم جديدOn the Surprising Efficiency of Committee-based Models
258
0
0.0
(
0
)
اسأل ChatGPT حول البحث
ﻻ يوجد ملخص باللغة العربية
Committee-based models, i.e., model ensembles or cascades, are underexplored in recent work on developing efficient models. While committee-based models themselves are not new, there lacks a systematic understanding of their efficiency in comparison with single models. To fill this gap, we conduct a comprehensive analysis of the efficiency of committee-based models. We find that committee-based models provide a complementary paradigm to achieve superior efficiency without tuning the architecture: even the most simplistic method for building ensembles or cascades from existing pre-trained networks can attain a significant speedup and higher accuracy over state-of-the-art single models, and also outperforms sophisticated neural architecture search methods (e.g., BigNAS). The superior efficiency of committee-based models holds true for several tasks, including image classification, video classification, and semantic segmentation, and various architecture families, such as EfficientNet, ResNet, MobileNetV2, and X3D.
قيم البحث
اقرأ أيضاً
How do humans learn to acquire a powerful, flexible and robust representation of objects? While much of this process remains unknown, it is clear that humans do not require millions of object labels. Excitingly, recent algorithmic advancements in sel
f-supervised learning now enable convolutional neural networks (CNNs) to learn useful visual object representations without supervised labels, too. In the light of this recent breakthrough, we here compare self-supervised networks to supervised models and human behaviour. We tested models on 15 generalisation datasets for which large-scale human behavioural data is available (130K highly controlled psychophysical trials). Surprisingly, current self-supervised CNNs share four key characteristics of their supervised counterparts: (1.) relatively poor noise robustness (with the notable exception of SimCLR), (2.) non-human category-level error patterns, (3.) non-human image-level error patterns (yet high similarity to supervised model errors) and (4.) a bias towards texture. Taken together, these results suggest that the strategies learned through todays supervised and self-supervised training objectives end up being surprisingly similar, but distant from human-like behaviour. That being said, we are clearly just at the beginning of what could be called a self-supervised revolution of machine vision, and we are hopeful that future self-supervised models behave differently from supervised ones, and---perhaps---more similar to robust human object recognition.
Purpose: Surgical task-based metrics (rather than entire procedure metrics) can be used to improve surgeon training and, ultimately, patient care through focused training interventions. Machine learning models to automatically recognize individual ta
sks or activities are needed to overcome the otherwise manual effort of video review. Traditionally, these models have been evaluated using frame-level accuracy. Here, we propose evaluating surgical activity recognition models by their effect on task-based efficiency metrics. In this way, we can determine when models have achieved adequate performance for providing surgeon feedback via metrics from individual tasks. Methods: We propose a new CNN-LSTM model, RP-Net-V2, to recognize the 12 steps of robotic-assisted radical prostatectomies (RARP). We evaluated our model both in terms of conventional methods (e.g. Jaccard Index, task boundary accuracy) as well as novel ways, such as the accuracy of efficiency metrics computed from instrument movements and system events. Results: Our proposed model achieves a Jaccard Index of 0.85 thereby outperforming previous models on robotic-assisted radical prostatectomies. Additionally, we show that metrics computed from tasks automatically identified using RP-Net-V2 correlate well with metrics from tasks labeled by clinical experts. Conclusions: We demonstrate that metrics-based evaluation of surgical activity recognition models is a viable approach to determine when models can be used to quantify surgical efficiencies. We believe this approach and our results illustrate the potential for fully automated, post-operative efficiency reports.
Instance segmentation models today are very accurate when trained on large annotated datasets, but collecting mask annotations at scale is prohibitively expensive. We address the partially supervised instance segmentation problem in which one can tra
in on (significantly cheaper) bounding boxes for all categories but use masks only for a subset of categories. In this work, we focus on a popular family of models which apply differentiable cropping to a feature map and predict a mask based on the resulting crop. Under this family, we study Mask R-CNN and discover that instead of its default strategy of training the mask-head with a combination of proposals and groundtruth boxes, training the mask-head with only groundtruth boxes dramatically improves its performance on novel classes. This training strategy also allows us to take advantage of alternative mask-head architectures, which we exploit by replacing the typical mask-head of 2-4 layers with significantly deeper off-the-shelf architectures (e.g. ResNet, Hourglass models). While many of these architectures perform similarly when trained in fully supervised mode, our main finding is that they can generalize to novel classes in dramatically different ways. We call this ability of mask-heads to generalize to unseen classes the strong mask generalization effect and show that without any specialty modules or losses, we can achieve state-of-the-art results in the partially supervised COCO instance segmentation benchmark. Finally, we demonstrate that our effect is general, holding across underlying detection methodologies (including anchor-based, anchor-free or no detector at all) and across different backbone networks. Code and pre-trained models are available at https://git.io/deepmac.
We investigate the sensitivity of the Frechet Inception Distance (FID) score to inconsistent and often incorrect implementations across different image processing libraries. FID score is widely used to evaluate generative models, but each FID impleme
ntation uses a different low-level image processing process. Image resizing functions in commonly-used deep learning libraries often introduce aliasing artifacts. We observe that numerous subtle choices need to be made for FID calculation and a lack of consistencies in these choices can lead to vastly different FID scores. In particular, we show that the following choices are significant: (1) selecting what image resizing library to use, (2) choosing what interpolation kernel to use, (3) what encoding to use when representing images. We additionally outline numerous common pitfalls that should be avoided and provide recommendations for computing the FID score accurately. We provide an easy-to-use optimized implementation of our proposed recommendations in the accompanying code.
Autonomous driving models should ideally be evaluated by deploying them on a fleet of physical vehicles in the real world. Unfortunately, this approach is not practical for the vast majority of researchers. An attractive alternative is to evaluate mo
dels offline, on a pre-collected validation dataset with ground truth annotation. In this paper, we investigate the relation between various online and offline metrics for evaluation of autonomous driving models. We find that offline prediction error is not necessarily correlated with driving quality, and two models with identical prediction error can differ dramatically in their driving performance. We show that the correlation of offline evaluation with driving quality can be significantly improved by selecting an appropriate validation dataset and suitable offline metrics. The supplementary video can be viewed at https://www.youtube.com/watch?v=P8K8Z-iF0cY
سجل دخول لتتمكن من نشر تعليقات
التعليقات
جاري جلب التعليقات
سجل دخول لتتمكن من متابعة معايير البحث التي قمت باختيارها
