Prasanna Sattigeri

Quantitative resolution of nanoparticle sizes using single particle inductively coupled plasma mass spectrometry with the K-means clustering algorithm

Sensitive and accurate characterization of nanoparticle size in aqueous matrices at environmentally relevant concentrations is still challenging for current nano-analysis techniques. Single particle inductively coupled plasma mass spectrometry (spICP-MS) is an emerging method to characterize the size distribution of nanoparticles and determine their concentrations. Herein for the first time the K-means clustering algorithm is applied to signal processing of spICP-MS raw data. Compared with currently used data processing approaches, the K-means algorithm improved discrimination of particle signals from background signals and provides a sophisticated, statistically based method to quantitatively resolve different size groups contained within a nanoparticle suspension. In tests with commercial Au nanoparticles (AuNPs), spICP-MS with the K-means clustering algorithm can quantitatively discriminate secondary “impurity-size nanoparticles,” present at fractions of less than 2% by mass, from primary-size nanoparticles with the minimum resolvable size difference between the primary and secondary nanoparticles at ∼20 nm. AuNP mixtures in which 80 nm particles act as the “primary size group” and 20 nm, 50 nm, or 100 nm particles act as the “impurity size group” were analyzed by spICP-MS, which reliably measured percentages of secondary impurity-size nanoparticles that are consistent with the expected experimentally determined values. Compared with dynamic light scattering (DLS), spICP-MS has remarkably better particle size resolution capability. We also demonstrated the size measurement advantage of spICP-MS over DLS for commercial CeO2 nanoparticles that are commonly used in the semiconductor industry, where quality control of the nanoparticle size distribution is critical for the wafer polishing process.

Supervised local sparse coding of sub-image features for image retrieval

The success of sparse representations in image modeling and recovery has motivated its use in computer vision applications. Image retrieval and classification tasks require extracting features that discriminate different image classes. State-of-the-art object recognition methods based on sparse coding use spatial pyramid features obtained from dense descriptors. In this paper, we develop a feature extraction method that uses multiple global/local features extracted from large overlapping regions of an image, which we refer to as sub-images. We propose a procedure for dictionary design and supervised local sparse coding of sub-image heterogeneous features. We perform image retrieval on the Microsoft Research Cambridge image dataset and show that the proposed features outperform the spatial pyramid features obtained using dense descriptors.

Learning dictionaries with graph embedding constraints

Several supervised, semi-supervised, and unsupervised machine learning schemes can be unified under the general framework of graph embedding. Incorporating graph embedding principles into sparse representation based learning schemes can provide an improved performance in several learning tasks. In this work, we propose a dictionary learning procedure for computing discriminative sparse codes that obey graph embedding constraints. In order to compute the graph-embedded sparse codes, we integrate a modified version of the sequential quadratic programming procedure with the feature sign search method. We demonstrate, using simulations with the AR face database, that the proposed approach performs better than several baseline methods in supervised and semi-supervised classification.

Implementation of a fast image coding and retrieval system using a GPU

Sparse coding of image patches is a compact but computationally expensive method of representing images. As part of our SenSIP consortium industry projects, we implement the Orthogonal Matching Pursuit algorithm using a single CUDA kernel on a GPU and sparse codes for image patches are obtained in parallel. Image-based “exact search” and “visually similar search” using the image patch sparse codes are performed. Results demonstrate large speed-up over CPU implementations and good retrieval performance is also achieved.

Transform domain features for ion-channel signal classification

The study of the behavior of ion-channels can provide significant information to detect metal ions and small organic molecules in solution. Discrimination of different analytes can be performed by extracting appropriate features from the ion-channel signals and using them for classification. In this paper, we consider features extracted from the Fourier, Wavelet and Walsh-Hadamard domain representations of the ion-channel signals. The proposed approach uses the power distribution information in the transform domains as features for discrimination. We compare the performance of all the three sets of features using support vector machines for classification of analytes and present the results. Results obtained show that the transform domain features achieve high classification rates in addition to high sensitivity and specificity rates.

Acquiring and Classifying Signals from Nanopores and Ion-Channels

The use of engineered nanopores as sensing elements for chemical and biological agents is a rapidly developing area. The distinct signatures of nanopore-nanoparticle lend themselves to statistical analysis. As a result, processing of signals from these sensors is attracting a lot of attention. In this paper we demonstrate a neural network approach to classify and interpret nanopore and ion-channel signals.

Transform domain features for ion-channel signal classification using support vector machines

The study of the behavior of ion-channels can provide significant information to detect metal ions and small organic molecules in solution. Discrimination of different analytes can be performed by extracting appropriate features from the ion-channel signals and using them for classification. In this paper, we consider features extracted from the Fourier, Wavelet and Walsh-Hadamard domain representations of the ion-channel signals. The proposed approach uses the power distribution information in the transform domains as features for discrimination. We compare the performance of all the three sets of features using support vector machines for classification of analytes and present the results. Results obtained show that the transform domain features achieve high classification rates in addition to high sensitivity and specificity rates.

A scalable feature learning and tag prediction framework for natural environment sounds

Building feature extraction approaches that can effectively characterize natural environment sounds is challenging due to the dynamic nature. In this paper, we develop a framework for feature extraction and obtaining semantic inferences from such data. In particular, we propose a new pooling strategy for deep architectures, that can preserve the temporal dynamics in the resulting representation. By constructing an ensemble of semantic embeddings, we employ an l1-reconstruction based prediction algorithm for estimating the relevant tags. We evaluate our approach on challenging environmental sound recognition datasets, and show that the proposed features outperform traditional spectral features.

Boosted dictionaries for image restoration based on sparse representations

Sparse representations using learned dictionaries have been successful in several image processing applications. However, using a single dictionary model in inverse problems may lead to instability in estimation. In this paper, we propose to perform image restoration using an ensemble of weak dictionaries that incorporate prior knowledge about the form of linear corruption. The dictionary learned in each round of the training procedure is optimized for the training examples having high reconstruction error in the previous round. The weak dictionaries are either obtained using a weighted K-Means or an example-selection approach. The final restored data is computed as a convex combination of data restored in individual rounds. Results with compressed recovery of standard images show that the proposed dictionaries result in a better performance compared to using a single dictionary obtained with a traditional alternating minimization approach.

Analyte detection using an ion-channel sensor array

Ion-channel sensors can be used for detecting small metal ions and organic molecules. The sensor consists of a chamber with a lipid bilayer hosting ion channels produced by protein insertion. These channels allow selective transport and produce a characteristic signal across the chamber for each analyte. A four chamber ion channel sensor array is built for accurate analyte detection. In this paper, we address the case in which non-uniform number of channels formed in each chamber. The power distribution information in the transform domain is used as features for each chamber signal. We employ support vector regression to estimate the number of channels inserted in each chamber and normalize the chamber signal features. The change observed in the normalized features of the chamber containing the analyte with respect to other chambers is used for detection. Results show high accuracy rates for detection of analyte using simulated data and experimental data.