An official website of the United States government
The concept of stimulus feature tuning isfundamental to neuroscience. Cortical neurons acquire their feature-tuning properties by learning from experience and using proxy signs of tentative features’ potential usefulness that come from the spatial and/or temporal context in which these features occur. According to this idea, local but ultimately behaviorally useful features should be the ones that are predictably related to other such features either preceding them in time or taking place side-by-side with them. Inspired by this idea, in this paper, deep neural networks are combined with Canonical Correlation Analysis (CCA) for feature extraction and the power of the features is demonstrated using unsupervised cross-modal prediction tasks. CCA is a multi-view feature extraction method that finds correlated features across multiple datasets (usually referred to as views or modalities). CCA finds linear transformations of each view such that the extracted principal components, or features, have a maximal mutual correlation. CCA is a linear method, and the features are computed by a weighted sum of each view's variables. Once the weights are learned, CCA can be applied to new examples and used for cross-modal prediction by inferring the target-view features of an example from its given variables in a source (query) view. To test the proposed method, it was applied to the unstructured CIFAR-100 dataset of 60,000 images categorized into 100 classes, which are further grouped into 20 superclasses and used to demonstrate the mining of image-tag correlations. CCA was performed on the outputs of three pre-trained CNNs: AlexNet, ResNet, and VGG. Taking advantage of the mutually correlated features extracted with CCA, a search for nearest neighbors was performed in the canonical subspace common to both the query and the target views to retrieve the most matching examples in the target view, which successfully predicted the superclass membership of the tested views without any supervised training.
more »
« less
Weak Target Detection in MIMO Radar via Beamspace Canonical Correlation
Reliable detection and accurate estimation of weak targets and their Doppler frequencies is a challenging problem in MIMO radar systems. Reflections from such targets are often overpowered by those from stronger nearby targets and clutter. Considering a 3-D data model where the coherent processing interval comprises multiple pulses, a novel weak target detection and estimation approach is proposed in this paper. The proposed method is based on creating partially overlapping spatial beams, and performing canonical correlation analysis (CCA) in the resulting beamspace. It is shown that if a target is present in the overlap sector, then its Doppler profile can be reliably estimated via beamspace CCA, even if hidden under much stronger interference from nearby targets and clutter. Numerical results are included to validate this theoretical claim, demonstrating that the proposed Beamspace Canonical Correlation (BCC) method yields considerable performance improvement over existing approaches.
more »
« less
- Award ID(s):
- 1807660
- PAR ID:
- 10176288
- Date Published:
- Journal Name:
- 2020 IEEE International Workshop on Sensor Array and Multichannel Signal Processing (SAM 2020)
- Page Range / eLocation ID:
- 1 to 5
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
A key challenge in 4G and emerging 5G systems is that of reliably detecting the uplink transmissions of users close to the edge between cells. These users are subject to significant signal attenuation due to path loss, and frequent hand-off from one cell to the other, making channel estimation very challenging. Even multiuser detection using base station cooperation often fails to detect such users, due to channel estimation errors and the sensitivity of multiuser detection to near-far power imbalance. Is it even possible to reliably decode the cell-edge users' signals under these circumstances? This paper shows, perhaps surprisingly, that with a suitable base station `interferometry' strategy, the cell-edge users' signals can be reliably decoded at low SNR under mild conditions. Exploiting the fact that cell-edge users' signals are weak but common to both base stations, while users close to a base station are unique to that base station, reliable detection is enabled by Canonical Correlation Analysis (CCA) - a machine learning technique that reliably estimates a common subspace, even in the presence of strong individual interference. Free from cell-center interference, the resulting mixture of cell-edge signals can then be unraveled using well-known algebraic signal processing techniques. Simulations demonstrate that the proposed detector achieves order of magnitude BER improvement compared to an `oracle' zero-forcing with successive interference cancellation that assumes perfect knowledge of all channels. The paper also includes proof of common subspace identifiability for the assumed generative model, which was curiously missing from the machine learning / CCA literature.more » « less
-
Canonical correlation analysis (CCA) is a technique for measuring the association between two multivariate data matrices. A regularized modification of canonical correlation analysis (RCCA) which imposes an [Formula: see text] penalty on the CCA coefficients is widely used in applications with high-dimensional data. One limitation of such regularization is that it ignores any data structure, treating all the features equally, which can be ill-suited for some applications. In this article we introduce several approaches to regularizing CCA that take the underlying data structure into account. In particular, the proposed group regularized canonical correlation analysis (GRCCA) is useful when the variables are correlated in groups. We illustrate some computational strategies to avoid excessive computations with regularized CCA in high dimensions. We demonstrate the application of these methods in our motivating application from neuroscience, as well as in a small simulation example.more » « less
-
NA (Ed.)Canonical correlation analysis (CCA) is a classic statistical method for discovering latent co-variation that underpins two or more observed random vectors. Several extensions and variations of CCA have been proposed that have strengthened our capabilities in terms of revealing common random factors from multiview datasets. In this work, we first revisit the most recent deterministic extensions of deep CCA and highlight the strengths and limitations of these state-of-the-art methods. Some methods allow trivial solutions, while others can miss weak common factors. Others overload the problem by also seeking to reveal what is {\em not common} among the views -- i.e., the private components that are needed to fully reconstruct each view. The latter tends to overload the problem and its computational and sample complexities. Aiming to improve upon these limitations, we design a novel and efficient formulation that alleviates some of the current restrictions. The main idea is to model the private components as {\em conditionally} independent given the common ones, which enables the proposed compact formulation. In addition, we also provide a sufficient condition for identifying the common random factors. Judicious experiments with synthetic and real datasets showcase the validity of our claims and the effectiveness of the proposed approach.more » « less
-
A visual stimulator plays an important role in a steady-state visual evoked potential (SSVEP)-based braincomputer interface (BCI). In conventional BCI studies, SSVEPs have been elicited by either a single stimulus whose flickering frequency varies across trials or multiple stimuli flickering at different frequencies simultaneously. It has been implicitly assumed that the SSVEPs generated by the single- and multiple-target stimulation methods are comparable. However, no study has directly compared their efficacy in eliciting SSVEPs. This study, therefore, performed a quantitative comparison of signal-to-noise ratio (SNR) and classification accuracy using 4-class SSVEPs generated by these two methods. The classification accuracy was estimated by three commonly-used target identification algorithms including calibration-free canonical correlation analysis (CCA)-based method and template-based methods with CCA- and task-related component analysis (TRCA)-based spatial filters. The results showed that the single-target stimulation method led to significantly higher SNR and classification accuracy than its multi-target counterpart.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1109/SAM48682.2020.9104284
