SVCCA: Singular Vector Canonical Correlation Analysis for Deep Learning Dynamics and Interpretability

By Maithra Raghu, Justin Gilmer, Jason Yosinski, Jascha Narain Sohl-Dickstein. 2017

906 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Model Architectures

Citations in the corpus

The Bottom-up Evolution of Representations in the Transformer: A Study with Machine Translation and Language Modeling Objectives (2019, methodology): “, 2018), which is an improved version of SVCCA (Raghu et al., 2017).”
MURAL: Multimodal, Multitask Retrieval Across Languages (2021, methodology): “We visualize multilingual text representations using Singular Value Canonical Correlation Analysis (SVCCA) (Raghu et al., 2017), which allows similarity scores to be computed between languages.”
High-Performance Large-Scale Image Recognition Without Normalization (2021, background): “However, being overly parsimonious with early stages (such as through aggressive downsampling) can hurt performance, since the model needs enough capacity to extract good local features (Raghu et al., 2017b).”
A Survey on Explainable Artificial Intelligence (XAI): Toward Medical XAI (2019, methodology): “Recently, Singular Vector Canonical Correlation Analysis (SVCCA) is suggested as a tool to analyze interpretability [68].”
On Learning the Geodesic Path for Incremental Learning (2021, methodology): “In this experiment, we explore the similarity notion between the current model and the old model using Canonical Correlation Analysis (CCA) in [5], as a tool to analyze the representation of deep models.”
Anatomy of Catastrophic Forgetting: Hidden Representations and Task Semantics (2020, methodology): “Representational similarity techniques take in a pair of neural network (hidden) layer representations and output a similarity score between 0 and 1 depicting how similar the representations are [21, 22, 38].”
Generative Feature Replay For Class-Incremental Learning (2020, methodology): “We use SVCCA [40] which first removes noise using singular value decomposition (SVD).”
Understanding intermediate layers using linear classifier probes (2016, background): “Refer to Figure 6 of Raghu et al. (2017b) for more details.”
Rapid Learning or Feature Reuse? Towards Understanding the Effectiveness of MAML (2019, methodology): “Table 3 presents CCA and CKA (another representational similarity measure, from [15]) similarities for MAML and ANIL models.”
Training Deeper Neural Machine Translation Models with Transparent Attention (2018, background): “Second, when a network is properly trained the lowest layers usually converge quickly, whereas the top-most layers take longer (Raghu et al., 2017).”
Identifying and Controlling Important Neurons in Neural Machine Translation (2018, methodology): “SVCCA Singular vector canonical correlation analysis ( SVCCA ) is a recent method for analyzing neural networks (Raghu et al., 2017).”
A Benchmark for Interpretability Methods in Deep Neural Networks (2018, methodology): “Other methods explore the role of neurons or activations in hidden layers of the network [24, 26, 23, 42], while others use high level concepts to explain prediction results [17].”
Emerging Cross-lingual Structure in Pretrained Language Models (2019, methodology): “Recent work introduces ways to measure the similarity of neural network activitation between different layers and different models (Laakso and Cottrell, 2000; Li et al., 2016; Raghu et al., 2017; Morcos et al., 2018; Wang et al., 2018).”
Investigating Multilingual NMT Representations at Scale (2019, methodology): “In this study we use Singular Value Canonical Correlation Analysis (SVCCA) (Raghu et al., 2017) as our investigation tool.”
Unsupervised Model Selection for Variational Disentangled Representation Learning (2019, background): “Other related approaches worth mentioning are the Canonical Correlation Analysis (CCA) and its modiﬁcations (Hardoon et al., 2004; Raghu et al., 2017; Morcos et al., 2018).”
Interpreting Black-Box Models: A Review on Explainable Artificial Intelligence (2023, methodology): “Several other methods, such as t-distributed stochastic neighbor embedding (t-SNE) and correlation-based singular vector canonical correlation analysis (SVCCA) [48], play a significant role in directing towards the sub-space of input for error-free predictions.”