اشترك بالحزمة الذهبية واحصل على وصول غير محدود شمرا أكاديميا
تسجيل مستخدم جديدHow Well do Feature Visualizations Support Causal Understanding of CNN Activations?
173
0
0.0
(
0
)
اسأل ChatGPT حول البحث
ﻻ يوجد ملخص باللغة العربية
One widely used approach towards understanding the inner workings of deep convolutional neural networks is to visualize unit responses via activation maximization. Feature visualizations via activation maximization are thought to provide humans with precise information about the image features that cause a unit to be activated. If this is indeed true, these synthetic images should enable humans to predict the effect of an intervention, such as whether occluding a certain patch of the image (say, a dogs head) changes a units activation. Here, we test this hypothesis by asking humans to predict which of two square occlusions causes a larger change to a units activation. Both a large-scale crowdsourced experiment and measurements with experts show that on average, the extremely activating feature visualizations by Olah et al. (2017) indeed help humans on this task ($67 pm 4%$ accuracy; baseline performance without any visualizations is $60 pm 3%$). However, they do not provide any significant advantage over other visualizations (such as e.g. dataset samples), which yield similar performance ($66 pm 3%$ to $67 pm 3%$ accuracy). Taken together, we propose an objective psychophysical task to quantify the benefit of unit-level interpretability methods for humans, and find no evidence that feature visualizations provide humans with better causal understanding than simple alternative visualizations.
قيم البحث
اقرأ أيضاً
Feature visualizations such as synthetic maximally activating images are a widely used explanation method to better understand the information processing of convolutional neural networks (CNNs). At the same time, there are concerns that these visuali
zations might not accurately represent CNNs inner workings. Here, we measure how much extremely activating images help humans to predict CNN activations. Using a well-controlled psychophysical paradigm, we compare the informativeness of synthetic images by Olah et al. (2017) with a simple baseline visualization, namely exemplary natural images that also strongly activate a specific feature map. Given either synthetic or natural reference images, human participants choose which of two query images leads to strong positive activation. The experiments are designed to maximize participants performance, and are the first to probe intermediate instead of final layer representations. We find that synthetic images indeed provide helpful information about feature map activations ($82pm4%$ accuracy; chance would be $50%$). However, natural images - originally intended as a baseline - outperform synthetic images by a wide margin ($92pm2%$). Additionally, participants are faster and more confident for natural images, whereas subjective impressions about the interpretability of the feature visualizations are mixed. The higher informativeness of natural images holds across most layers, for both expert and lay participants as well as for hand- and randomly-picked feature visualizations. Even if only a single reference image is given, synthetic images provide less information than natural images ($65pm5%$ vs. $73pm4%$). In summary, synthetic images from a popular feature visualization method are significantly less informative for assessing CNN activations than natural images. We argue that visualization methods should improve over this baseline.
Analysts often make visual causal inferences about possible data-generating models. However, visual analytics (VA) software tends to leave these models implicit in the mind of the analyst, which casts doubt on the statistical validity of informal vis
ual insights. We formally evaluate the quality of causal inferences from visualizations by adopting causal support -- a Bayesian cognition model that learns the probability of alternative causal explanations given some data -- as a normative benchmark for causal inferences. We contribute two experiments assessing how well crowdworkers can detect (1) a treatment effect and (2) a confounding relationship. We find that chart users causal inferences tend to be insensitive to sample size such that they deviate from our normative benchmark. While interactively cross-filtering data in visualizations can improve sensitivity, on average users do not perform reliably better with common visualizations than they do with textual contingency tables. These experiments demonstrate the utility of causal support as an evaluation framework for inferences in VA and point to opportunities to make analysts mental models more explicit in VA software.
State-of-the-art summarization systems are trained and evaluated on massive datasets scraped from the web. Despite their prevalence, we know very little about the underlying characteristics (data noise, summarization complexity, etc.) of these datase
ts, and how these affect system performance and the reliability of automatic metrics like ROUGE. In this study, we manually analyze 600 samples from three popular summarization datasets. Our study is driven by a six-class typology which captures different noise types (missing facts, entities) and degrees of summarization difficulty (extractive, abstractive). We follow with a thorough analysis of 27 state-of-the-art summarization models and 5 popular metrics, and report our key insights: (1) Datasets have distinct data quality and complexity distributions, which can be traced back to their collection process. (2) The performance of models and reliability of metrics is dependent on sample complexity. (3) Faithful summaries often receive low scores because of the poor diversity of references. We release the code, annotated data and model outputs.
The power of neural networks lies in their ability to generalize to unseen data, yet the underlying reasons for this phenomenon remain elusive. Numerous rigorous attempts have been made to explain generalization, but available bounds are still quite
loose, and analysis does not always lead to true understanding. The goal of this work is to make generalization more intuitive. Using visualization methods, we discuss the mystery of generalization, the geometry of loss landscapes, and how the curse (or, rather, the blessing) of dimensionality causes optimizers to settle into minima that generalize well.
Sign language translation (SLT) aims to interpret sign video sequences into text-based natural language sentences. Sign videos consist of continuous sequences of sign gestures with no clear boundaries in between. Existing SLT models usually represent
sign visual features in a frame-wise manner so as to avoid needing to explicitly segmenting the videos into isolated signs. However, these methods neglect the temporal information of signs and lead to substantial ambiguity in translation. In this paper, we explore the temporal semantic structures of signvideos to learn more discriminative features. To this end, we first present a novel sign video segment representation which takes into account multiple temporal granularities, thus alleviating the need for accurate video segmentation. Taking advantage of the proposed segment representation, we develop a novel hierarchical sign video feature learning method via a temporal semantic pyramid network, called TSPNet. Specifically, TSPNet introduces an inter-scale attention to evaluate and enhance local semantic consistency of sign segments and an intra-scale attention to resolve semantic ambiguity by using non-local video context. Experiments show that our TSPNet outperforms the state-of-the-art with significant improvements on the BLEU score (from 9.58 to 13.41) and ROUGE score (from 31.80 to 34.96)on the largest commonly-used SLT dataset. Our implementation is available at https://github.com/verashira/TSPNet.
الأسئلة المقترحة
سجل دخول لتتمكن من نشر تعليقات
التعليقات
جاري جلب التعليقات
سجل دخول لتتمكن من متابعة معايير البحث التي قمت باختيارها
