Zhifei Zhang

Age Progression/Regression by Conditional Adversarial Autoencoder

If I provide you a face image of mine (without telling you the actual age when I took the picture) and a large amount of face images that I crawled (containing labeled faces of different ages but not necessarily paired), can you show me what I would look like when I am 80 or what I was like when I was 5? The answer is probably a No. Most existing face aging works attempt to learn the transformation between age groups and thus would require the paired samples as well as the labeled query image. In this paper, we look at the problem from a generative modeling perspective such that no paired samples is required. In addition, given an unlabeled image, the generative model can directly produce the image with desired age attribute. We propose a conditional adversarial autoencoder (CAAE) that learns a face manifold, traversing on which smooth age progression and regression can be realized simultaneously. In CAAE, the face is first mapped to a latent vector through a convolutional encoder, and then the vector is projected to the face manifold conditional on age through a deconvolutional generator. The latent vector preserves personalized face features (i.e., personality) and the age condition controls progression vs. regression. Two adversarial networks are imposed on the encoder and generator, respectively, forcing to generate more photo-realistic faces. Experimental results demonstrate the appealing performance and flexibility of the proposed framework by comparing with the state-of-the-art and ground truth.

Robust coupling in space of sparse codes for multi-view recognition

Classical dictionary learning algorithms that rely on a single source of information have been successfully used for classification tasks. Additionally, the exploitation of multiple sources has shown to be advantageous in challenging real-world situations. We propose a new framework to exploit robust modality fusion in classification in order to achieve better classification performance than single source methods. Multimodal learning is able to leverage any correlations between sensor modalities found in the data. We propose a new bilevel optimization, referred to as (MCJWDL). We perform supervised dictionary learning while forcing a coupling between the resulting sparse codes from different sources of information. Extensive experiments demonstrate that MCJWDL outperforms state-of-the-art sparse representation and dictionary learning approaches for the multi-view object and multi-view action recognition.

Multiple event analysis for large-scale power systems through cluster-based sparse coding

Accurate event analysis in real time is of paramount importance for high-fidelity situational awareness such that proper actions can take place before any isolated faults escalate to cascading blackouts. For large-scale power systems, due to the large intra-class variance and inter-class similarity, the nonlinear nature of the system, and the large dynamic range of the event scale, multi-event analysis presents an intriguing problem. Existing approaches are limited to detecting only single or double events or a specified event type. Although some previous works can well distinguish multiple events in small-scale power systems, the performance tends to degrade dramatically in large-scale systems. In this paper, we focus on multiple event detection, recognition, and temporal localization in large-scale power systems. We discover that there always exist groups of buses whose reaction to each event shows high degree similarity, and the group membership generally remains the same regardless of the type of event(s). We further verify that this reaction to multiple events can be approximated as a linear combination of reactions to each constituent event. Based on these findings, we propose a novel method, referred to as cluster-based sparse coding (CSC), to extract all the underlying single events involved in a multi-event scenario. Experimental results based on simulated large-scale system model (i.e., NPCC) show that the proposed CSC algorithm presents high detection and recognition rate with low false alarms.

Early mastitis diagnosis through topological analysis of biosignals from low-voltage alternate current electrokinetics.

Mastitis is the most economically important disease of dairy cows worldwide, and it constantly plagues the dairy industry. A reliable biosensing method is thus imperative to detect this disease at its early stage and accurately identify the pathogen concentration level in order to better control the disease and consequently improve the quality of milk. Recent research indicates that shorter assay time and/or higher sensitivity can be achieved by integrating alternate current electrokinetics (ACEK) with biosensing. However, most existing ACEK devices use voltage levels around 10V at the risk of electrochemical reactions because a lower voltage may not effectively trigger the ACEK effect. Currently, there are no related works that can efficiently tackle the dilemma between avoiding electrochemical reaction and accelerating assay process. This paper adopts low-voltage (40~135mV) ACEK, which is safe but yields ambiguous biosignals within a short assay time, presenting great challenge to high-fidelity identification of pathogen concentration levels. This paper makes two distinctive contributions to the field of biosignal analysis. First, moving away from the traditional signal analysis in the time or spectral domain, we exploit the possibility of representing the biosignal through topological analysis that would reveal the intrinsic topological structure of point clouds generated from the biosignal. Second, in order to tackle another common challenge of biosignal analysis, i.e., limited sample size, we propose a so-called Gaussian-based decision tree (GDT), which can efficiently classify the biosignals even when the sample size is extremely small. Experimental results on the classification of five pathogen concentration levels using only 10 samples taken under various voltage levels demonstrate the robustness of the topological features as well as the advantage of GDT over some other conventional classifiers in handling small dataset. Our method reduces the voltage of ACEK to a safe level and still yields high-fidelity results in a short time.

Dictionary Reduction: Automatic Compact Dictionary Learning for Classification

A complete and discriminative dictionary can achieve superior performance. However, it also consumes extra processing time and memory, especially for large datasets. Most existing compact dictionary learning methods need to set the dictionary size manually, therefore an appropriate dictionary size is usually obtained in an exhaustive search manner. How to automatically learn a compact dictionary with high fidelity is still an open challenge. We propose an automatic compact dictionary learning (ACDL) method which can guarantee a more compact and discriminative dictionary while at the same time maintaining the state-of-the-art classification performance. We incorporate two innovative components in the formulation of the dictionary learning algorithm. First, an indicator function is introduced that automatically removes highly correlated dictionary atoms with weak discrimination capacity. Second, two additional constraints, namely, the sum-to-one and the non-negative constraints are imposed on the sparse coefficients. On one hand, this achieves the same functionality as the \(L_2\)-normalization on the raw data to maintain a stable sparsity threshold. On the other hand, this effectively preserves the geometric structure of the raw data which would be otherwise destroyed by the \(L_2\)-normalization. Extensive evaluations have shown that the preservation of geometric structure of the raw data plays an important role in achieving high classification performance with smallest dictionary size. Experimental results conducted on four recognition problems demonstrate the proposed ACDL can achieve competitive classification performance using a drastically reduced dictionary (https://github.com/susanqq/ACDL.git).

Derivative Delay Embedding: Online Modeling of Streaming Time Series

The staggering amount of streaming time series coming from the real world calls for more efficient and effective online modeling solution. For time series modeling, most existing works make some unrealistic assumptions such as the input data is of fixed length or well aligned, which requires extra effort on segmentation or normalization of the raw streaming data. Although some literature claim their approaches to be invariant to data length and misalignment, they are too time-consuming to model a streaming time series in an online manner. We propose a novel and more practical online modeling and classification scheme, DDE-MGM, which does not make any assumptions on the time series while maintaining high efficiency and state-of-the-art performance. The derivative delay embedding (DDE) is developed to incrementally transform time series to the embedding space, where the intrinsic characteristics of data is preserved as recursive patterns regardless of the stream length and misalignment. Then, a non-parametric Markov geographic model (MGM) is proposed to both model and classify the pattern in an online manner. Experimental results demonstrate the effectiveness and superior classification accuracy of the proposed DDE-MGM in an online setting as compared to the state-of-the-art.

Multiple Event Detection and Recognition for Large-Scale Power Systems Through Cluster-Based Sparse Coding

Accurate event analysis in real time is of paramount importance for high-fidelity situational awareness such that proper actions can take place before any isolated faults escalate to cascading blackouts. Existing approaches are limited to detect only single or double events or a specified event type. Although some previous works can well distinguish multiple events in small-scale systems, the performance tends to degrade dramatically in large-scale systems. In this paper, we focus on multiple event detection, recognition, and temporal localization in large-scale power systems. We discover that there always exist “regions” where the reaction of all buses to certain event within each region demonstrates high degree similarity, and that the boundary of the “regions” generally remains the same regardless of the type of event(s). We further verify that, within each region, this reaction to multiple events can be approximated as a linear combination of reactions to each constituent event. Based on these findings, we propose a novel method, referred to as cluster-based sparse coding (CSC), to extract all the underlying single events involved in a multievent scenario. Multiple events of three typical disturbances (e.g., generator trip, line trip, and load shedding) can be detected and recognized. Specifically, the CSC algorithm can effectively distinguish line trip events from oscillation, which has been a very challenging task for event analysis. Experimental results based on simulated large-scale system model (i.e., NPCC) show that the proposed CSC algorithm presents high detection and recognition rate with low false alarms.

Topological Analysis and Gaussian Decision Tree: Effective Representation and Classification of Biosignals of Small Sample Size

Goal: Bucking the trend of big data, in microdevice engineering, small sample size is common, especially when the device is still at the proof-of-concept stage. The small sample size, small interclass variation, and large intraclass variation, have brought biosignal analysis new challenges. Novel representation and classification approaches need to be developed to effectively recognize targets of interests with the absence of a large training set. Methods: Moving away from the traditional signal analysis in the spatiotemporal domain, we exploit the biosignal representation in the topological domain that would reveal the intrinsic structure of point clouds generated from the biosignal. Additionally, we propose a Gaussian-based decision tree (GDT), which can efficiently classify the biosignals even when the sample size is extremely small. Results: This study is motivated by the application of mastitis detection using low-voltage alternating current electrokinetics (ACEK) where five categories of bisignals need to be recognized with only two samples in each class. Experimental results demonstrate the robustness of the topological features as well as the advantage of GDT over some conventional classifiers in handling small dataset. Conclusion: Our method reduces the voltage of ACEK to a safe level and still yields high-fidelity results with a short assay time. Significance: This paper makes two distinctive contributions to the field of biosignal analysis, including performing signal processing in the topological domain and handling extremely small dataset. Currently, there have been no related works that can efficiently tackle the dilemma between avoiding electrochemical reaction and accelerating assay process using ACEK.