RNNbow: Visualizing Learning Via Backpropagation Gradients in RNNs

Cashman, Dylan; Patterson, Geneviève; Mosca, Ab; Watts, Nathan; Robinson, Shannon; Chang, Remco

doi:10.1109/mcg.2018.2878902

Cited by 42 publications

(37 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…An alternative to visualizing what a model has learned is visualizing how it is learning. RNNbow by Cashman et al [5] shows the gradient flow during backpropagation training in RNNs to visualize how the network is learning.…”

Section: Related Workmentioning

confidence: 99%

Seq2seq-Vis: A Visual Debugging Tool for Sequence-to-Sequence Models

Strobelt

Gehrmann²,

Behrisch³

et al. 2019

IEEE Trans. Visual. Comput. Graphics

198

125

View full text Add to dashboard Cite

Start entering some encoder sentence (enter triggers request)... our tool helps to find errors in seq2seq models using visual analysis methods .Enc words: Attention: topK:our tool helps to find errors in seq2seq models using visual analysis methods .unser werkzeug hil , fehler in modellen zu finden , die mit visuellen analysen .unser werkzeug hil , fehler in modellen zu finden , die mit visuellen analysen der .unsere instrument dabei dabei fehlern zu der modelle anzuwenden entdecken mit mit mittels visueller von das tool hilfreich dazu abwei chungen bei den einzusetzen suchen die indem visuell der belegen werden unserem hilfsmittel ist zu für form , mit verschaffen mittels mittels visuellen vi suali si erung auswertung geprägt zu wir werkzeuge helfen es etwas auf die anhand für geben und um von analyse zu lernen pivot change: word attn compare: sentence swap: unser unsere werkzeug instrument tool hil , dabei es dazu fehler , fehler in zu fehler zu in finden in finden modellen modelle in in zu zu modellen finden finden modellen zu modellen , , , finden die die , mit mittels mit visuellen visueller visueller visuellen analysen analysen der von der von visuellen . visuellen analyse analysen analyse . zu . . show: edges nodes our tool helps to find errors in seq2seq models using visual analysis methods . show: src tgt highlight: -1 0 +1 and around the world , satellites and warning systems are saving lives in areas such as bangladesh . these are the two pictures taken of garment factories in province and garment factories in india . i would love to talk about my astronomy , but i suspect that the number of people who are interested in transfer in atmospheres and polarization of light in jupiter 's upper atmosphere are the number of people who 'd fit in a bus shelter . if a neutrino had a brain , which it evolved in ancestors , it would say that rocks really do consist of empty space . i would love to talk about my astronomy , but i suspect that the number of people who are interested in transfer in atmospheres and polarization of light in jupiter 's upper atmosphere are the number of people who 'd fit in a bus shelter . most of those individuals had spent most of their lives in hospitals . this is a long time ago . that 's year by year . this comes from our friends at i 'll get an esl class in learning " it 's raining , it 's pouring . " you send one blessed email to whomever you 're thinking of at she for jobs down in province in the south . science columnist lee describes a remarkable project at divide , antarctica , where a hardy team are drilling into ice to extract vital data on our changing climate . that technology will be used on animals . i work in homes , largely . so for example , there was one study that was done in a population of jews in new yor...

~~show abstract~~

Section: Related Workmentioning
confidence: 99%

Seq2seq-Vis: A Visual Debugging Tool for Sequence-to-Sequence Models
Strobelt
¹
,
Gehrmann²,
Behrisch³
et al. 2019
IEEE Trans. Visual. Comput. Graphics
198
1
125
0
View full text Add to dashboard Cite

~~show abstract~~

“…Regarding the former, various visual analytic approaches have been proposed for convolutional neural networks mainly computer vision domains [2,6,12,13,19,34] and RNNs in NLP domains [5,11,17,23,24]. Visual analytic approaches have also been integrated with other advanced neural network architectures, such as generative adversarial networks [9,30], deep reinforcement learning [29].…”

Section: Related Workmentioning
confidence: 99%

SANVis: Visual Analytics for Understanding Self-Attention Networks
Park¹,
Choo²,
Na³
et al. 2019
2019 IEEE Visualization Conference (VIS)
22
1
26
1
View full text Add to dashboard Cite

Figure 1: Overview of SANVis. (A) The network view displays multiple attention patterns for each layer according to three type of visualization options: (A-1) the attention piling option, (A-2) the Sankey diagram option, and (A-3) the small multiples option. (A-4) The bar chart shows the average attention weights for all heads (each colored with its corresponding hue) per each layer. (B) The HeadLens view helps the user analyze what the attention head learned by showing representative words and by providing statistical information of part-of-speech tags and positions. ABSTRACTAttention networks, a deep neural network architecture inspired by humans' attention mechanism, have seen significant success in image captioning, machine translation, and many other applications. Recently, they have been further evolved into an advanced approach called multi-head self-attention networks, which can encode a set of input vectors, e.g., word vectors in a sentence, into another set of vectors. Such encoding aims at simultaneously capturing diverse syntactic and semantic features within a set, each of which corresponds to a particular attention head, forming altogether multi-head attention. Meanwhile, the increased model complexity prevents users from easily understanding and manipulating the inner workings of models. To tackle the challenges, we present a visual analytics system called SANVis, which helps users understand the behaviors and the characteristics of multi-head self-attention networks. Using a state-of-the-art self-attention model called Transformer, we demonstrate usage scenarios of SANVis in machine translation tasks. Our system is available at http://short.sanvis.org.

~~show abstract~~

“…While the calculations used in machine learning models can be excessively complicated, endemic properties of models that cause poor predictions can sometimes be diagnosed visually relatively easily. RNNBow is a tool that uses intermediate training data to visually reveal issues with gradient flow during the training process of a recurrent neural network [CPMC17]. ] provide an interactive analysis method to debug and analyze weather forecast models based on their confidence estimates.…”

Section: Modeling In Visual Analyticsmentioning
confidence: 99%

A User‐based Visual Analytics Workflow for Exploratory Model Analysis
Cashman
¹
,
Humayoun
²
,
Heimerl
³

et al. 2019
Computer Graphics Forum
Self Cite
40
0
42
0
View full text Add to dashboard Cite

Many visual analytics systems allow users to interact with machine learning models towards the goals of data exploration and insight generation on a given dataset. However, in some situations, insights may be less important than the production of an accurate predictive model for future use. In that case, users are more interested in generating of diverse and robust predictive models, verifying their performance on holdout data, and selecting the most suitable model for their usage scenario. In this paper, we consider the concept of Exploratory Model Analysis (EMA), which is defined as the process of discovering and selecting relevant models that can be used to make predictions on a data source. We delineate the differences between EMA and the well‐known term exploratory data analysis in terms of the desired outcome of the analytic process: insights into the data or a set of deployable models. The contributions of this work are a visual analytics system workflow for EMA, a user study, and two use cases validating the effectiveness of the workflow. We found that our system workflow enabled users to generate complex models, to assess them for various qualities, and to select the most relevant model for their task.

~~show abstract~~

RNNbow: Visualizing Learning Via Backpropagation Gradients in RNNs

Cited by 42 publications

References 13 publications

Seq2seq-Vis: A Visual Debugging Tool for Sequence-to-Sequence Models

Seq2seq-Vis: A Visual Debugging Tool for Sequence-to-Sequence Models

SANVis: Visual Analytics for Understanding Self-Attention Networks

A User‐based Visual Analytics Workflow for Exploratory Model Analysis

Contact Info

Product

Resources

About