News

MCL Research on Point Cloud Classification and Segmentation

Recently, Professor Kuo and his students at MCL proposed a new machine learning methodology called successive subspace learning (SSL). The methodology has been widely adopted in MCL to solve image processing and computer vision problems. In 3D domain, we have observed a great success in point cloud classification task. In the PointHop paper, we develop an explainable machine learning method for point cloud classification. The classification baseline is composed by four PointHop units, we construct the local-to-global attribute building process and use saab transform to control the dimension growth in each unit. We compare the test performance on ModelNet40 with the state-of-the-art methods, our method obtains comparable performance with the others while demands much less training time. For instance, PointNet costs about 5 hours to train, while ours only takes 20 minutes to train on the same dataset. The advantages of the methodology are very clear: interpretable and much less computation complexity.

The success in point cloud classification encourages us to go deeper in 3D domain. Therefore, we further look at the segmentation task which needs to assign label to each point in the point cloud. Referring to the common image segmentation network, we use the point cloud classification baseline as an encoder and add a decoder to complete segmentation. After building local neighboring regions and extracting local attributes from neighboring points in the encoder, the features are interpolated back to finest scale layer by layer with skip connections between same scales in the decoder. Also, saab transform is adopted between layers as feedfoward convolution to control the rapid growth of the feature dimension.

Our method also has the advantage of task-agnostic ability. Specifically, by learning the parameters in a one-pass manner, our [...]

By |November 11th, 2019|News|Comments Off on MCL Research on Point Cloud Classification and Segmentation|

MCL Research on Successive Subspace Learning

Subspace methods have been widely used in signal/image processing, pattern recognition, computer vision, etc.   One may use a subspace to denote the feature space of a certain object class, (e.g., the subspace of the dog object class) or the dominant feature space by dropping less important features (e.g., the subspace obtained via principal component analysis or PCA). The subspace representation offers a powerful tool for signal analysis, modeling and processing. Subspace learning is to find subspace models for concise data representation and accurate decision making based on training samples.

Most existing subspace methods are conducted in a single stage. We may ask whether there is an advantage to perform subspace learning in multiple stages. Research on generalizing from one-stage subspace learning to multi-stage subspace learning is rare. Two PCA stages are cascaded in the PCAnet, which provides an empirical solution to multi-stage subspace learning. Little research on this topic may be attributed to the fact that a straightforward cascade of linear multi-stage subspace methods, which can be expressed as the product of a sequence of matrices, is equivalent to a linear one-stage subspace method. The advantage of linear multi-stage subspace methods may not be obvious from this viewpoint.

Yet, multi-stage subspace learning may be worthwhile under the following two conditions. First, the input subspace is not fixed but growing from one stage to the other. For example, we can take the union of a pixel and its eight nearest neighbors to form an input space in the first stage. Afterward, we enlarge the neighborhood of the center pixel from 3×3 to 5×5 in the second stage.  Clearly, the first input space is a proper subset of the second input space. By generalizing it to multiple stages, [...]

By |November 3rd, 2019|News|Comments Off on MCL Research on Successive Subspace Learning|

Welcome New MCL Member – Eric Huang

We are so happy to welcome a new undergraduate member of MCL, Eric Huang. Here is an interview with Eric:

Could you briefly introduce yourself and your research interests?

Hi, my name is Eric Huang and I’m a current senior studying computer science at USC. I’ve been interested in research ever since I started college and the MCL is a lab I’m honored to have been able to join. I previously worked in biomedical and mechanical engineering labs, but machine learning is the field I am most interested in. My research focus is on facial classification and verification, particularly in studying how to streamline the training process. But I think that every new development in AI is incredible and I hope to be able to learn and understand the many aspects.

What is your impression about MCL and USC?

My impression of USC after three years is that the school is amazing. There are countless intelligent students and professors that form a productive and supportive network. I see the MCL as the pinnacle of what USC can achieve. Everyone in the MCL is highly driven, intelligent, and passionate and I hope to live up to the same standard.

What is your future expectation and plan in MCL?

My expectation and plan in the MCL are to learn, grow, and hopefully contribute. I plan to study hard and learn everything I can in order to be able to meaningfully help out. I expect to learn a lot from the lab members and to work hard supporting them. And I hope that I am able to form lasting connections with all the talented people I get to work with.

By |October 27th, 2019|News|Comments Off on Welcome New MCL Member – Eric Huang|

Professor Kuo visited Nanjing, Hefei and Zhuzhou

MCL Director, Professor C.-C. Jay Kuo, visited Nanjing, Hefei and Zhuzhou in the week of October 7-13.

Professor Kuo’s visit to Nanjing University on October 8 (Tuesday) was invited and hosted by Professor Zhi-Hua Zhou, Dean of School of Artificial Intelligence and Head of Department of Computer Science and Technology, Nanjing University. Professor Kuo delivered a seminar on “From Feedforward-Designed Convolutional Neural Networks (FF-CNNs) to Successive Subspace Learning (SSL)” as part of the CSAI Distinguished Lecture Series of Nanjing University. A photo of Professor Kuo and Professor Zhou is provided.

Professor Kuo’s visit to University of Science and Technology of China on October 9 (Wednesday) was invited and co-hosted by Professor Feng Wu, Professor Houqiang Li and Professor Qibin Sun. Professor Kuo delivered a seminar on his recent work on Success Subspace Learning (SSL) to faculty and students.

At the last stop of his trip, Professor Kuo attended the Chinese Conference on Biometric Recognition (CCBR) from October 12 and 13 in Zhuzhou, Hunan, China. He was a keynote speaker of this conference and delivered a lecture on “Towards Effective and Explainable Biometrics”. The abstract of his keynote is given below.

“Deep learning provides state-of-the-art biometrics solutions when training and testing data share similar distributions and the number of training samples is sufficiently larger. The deep-learning-based solutions are mathematically intractable due to the non-convex optimization nature. Furthermore, their robustness is a main concern. To search for effective and explainable biometrics solutions is challenging yet essential. In this talk, I will present a path towards this direction and provide preliminary results using face recognition as an example. Instead of treating computational neurons as hidden units whose parameters are determined by end-to-end optimization, we interpret computational neurons as dimension reduction units, [...]

By |October 20th, 2019|News|Comments Off on Professor Kuo visited Nanjing, Hefei and Zhuzhou|

Welcome New MCL Member Xiou Ge

We are so happy to welcome a new graduate member of MCL, Xiou Ge. Here is an interview with Xiou:

Could you briefly introduce yourself and your research interests?

My name is Xiou Ge. I’m a first year PhD student in Electrical Engineering. I’m from Harbin, China. After completing my middle school education in China, I was awarded a scholarship and spend my high school years in Singapore. I obtained my bachelor’s degree with highest honors and the master’s degree, both in Electrical Engineering from the University of Illinois at Urbana-Champaign. Previously, I interned at Apple in Cupertino and worked on EDA tool development, and IBM T.J. Watson Research Center in Yorktown Heights and did computational creativity research. My current research interests include artificial intelligence, machine learning, and computer vision.

What is your impression about MCL and USC?

Everyone I met in MCL has been very nice and willing to help each other. The atmosphere is very relaxed and people interact with each other just like family members. Yet everybody works very hard and I can feel the energy which motivates me to work harder. I think MCL has the conducive environment for me to become a successful graduate student. USC offers a very different college experience from my previous one. Located in the middle of downtown LA, the school offers plenty of opportunities for students to experience and learn from the outside world. Collaborations with researchers from other institutions and other parts of the world are frequent and convenient. The school also makes substantial investment in education and allows students to learn from and do research with world-class faculty and experts in different research fields. 

What is your future expectation and plan in [...]

By |October 13th, 2019|News|Comments Off on Welcome New MCL Member Xiou Ge|

Welcome New MCL Member – Hong-Shuo Chen

We are so happy to welcome a new undergraduate member of MCL, Hong-Shuo Chen. Here is an interview with Hong-Shuo:

1. Could you briefly introduce yourself and your research interests?

I am Hong-Shuo Chen, and my English name is Max. Before coming to USC, I got my bachelor degree from Electrical Engineering and Computer Science at National Chiao Tung University in Taiwan. I like coding, math and everything about engineering. It gives me great pleasure to come to MCL as a Ph.D. student. My research interests are image segmentation and texture analysis. The world of the computer vision is very broad that I want to explore more in this field.

2. What is your impression about MCL and USC?

USC is a top university of this world and MCL is really a cooperative and strong lab. I like this beautiful campus and really enjoy doing research in MCL. Professor Kuo is really a good mentor. He takes care of all the members in the lab and also has profound knowledge and experience. I really appreciate having this opportunity to learn and study in MCL.

3. What is your future expectation and plan in MCL?

With the guidance of Professor Kuo, I believe I can dive into the world of computer vision deeply and thoroughly, and become expert in the signal processing and machine learning. In the future, I hope I can become a professional engineer, solve problems of this world independently and make some contribution to this society.

By |October 6th, 2019|News|Comments Off on Welcome New MCL Member – Hong-Shuo Chen|

Welcome New MCL Member – Hongyu Fu

We are so happy to welcome a new graduate member of MCL, Hongyu Fu. Here is an interview with Hongyu:

1. Could you briefly introduce yourself and your research interests?
My name is Hongyu Fu, before becoming a PhD student in USC, I got my bachelor’s degree in electrical engineering from Peking University. My past research experience focuses mostly on semiconductor device physics and circuits, while exploring device and circuit based neuromorphic computing and machine learning topics, I have heard more and more about computer vision, AI and machine learning, which are always my keen interest that I haven’t got chances to learn before. Therefore, I really appreciate Prof. Kuo for giving me this opportunity to study in MCL and explore more in this exciting area.
 
2. What is your impression about MCL and USC?
I really love the beautiful campus, friendly atmosphere and enjoy the convenient facilities of USC. We have a fight-on spirit as Trojans, which makes USC a great network and everyone in this network passionate and motivated. MCL is a large and efficient group with a hard-working spirit and intelligent students with solid skills and deep knowledge under the supervision of Prof. Kuo, who is a very nice and responsible advisor profound in knowledge and research.
 

3. What is your future expectation and plan in MCL?
I will definitely work hard, learn solid skills of math, problems solving and research and hope to contribute to MCL in the future. I wish that in the near future, the MCL group and myself would have a solid standing in the machine learning and computer vision community and contribute to the improvement of this field.

By |September 29th, 2019|News|Comments Off on Welcome New MCL Member – Hongyu Fu|

MCL Research on Robot Learning

Title: Robot Learning via Human Adversarial Games
Author: Jiali Duan

Much work in robotics has focused on “humanin-the-loop” learning techniques that improve the efficiency of the learning process. However, these algorithms have made the strong assumption of a cooperating human supervisor that assists the robot. In reality, human observers tend to also act in an adversarial manner towards deployed robotic systems. We show that this can in fact improve the robustness of the learned models by proposing a physical framework that leverages perturbations applied by a human adversary, guiding the robot towards more robust models. In a manipulation task, we show that grasping success improves significantly when the robot trains with a human adversary as compared to training in a self-supervised manner. We validate our approach in a self-brewed simulator for human-robot interaction. Our work has been selected as Best Paper Finalist for IROS 2019 and more details can be found at: https://arxiv.org/abs/1903.00636.

Before training:

 

After training:

By |September 22nd, 2019|News|Comments Off on MCL Research on Robot Learning|

Congratulations to Yueru Chen for Passing Her PhD Defense

The thesis is entitled “Object Classification Based on Neural-Network-Inspired Image Transforms”.

Abstract of the thesis:

Convolutional neural networks (CNNs) have recently demonstrated impressive performance in image classification and change the way building feature extractors from carefully handcrafted design to automatically deep learned from a large labeled dataset. However, a great majority of current CNN literature are application-oriented, and there is no clear understanding and theoretical foundation to explain the outstanding performance and indicate the way to improve. In this thesis, we focus on solving the image classification problem-based on the neural-network-inspired transforms.

Being motivated by the multilayer RECOS (REctified-COrrelations on a Sphere) transform, two data-driven signal transforms are proposed, called the “Subspace approximation with augmented kernels” (Saak) transform and “Subspace approximation with adjusted bias” (Saab) transform corresponding to each Convolutional layers in CNNs. Based on the Saak transform, We firstly proposed an efficient, scalable and robust approach to the handwritten digits recognition problem. Next, we also develop an ensemble method using Saab transform to solve the image classification problem. The ensemble method fuses the output decision vectors of Saab-transform-based decision system. To enhance the performance of the ensemble system, it is critical to increasing the diversity of FF-CNN models. To achieve this objective, we introduce diversities by adopting three strategies: 1) different parameter settings in convolutional layers, 2) flexible feature subsets fed into the Fully-connected (FC) layers, and 3) multiple image embeddings of the same input source.  We also extend our ensemble method to semi-supervised learning. Since unlabeled data may not always enhance semi-supervised learning, we define an effective quality score and use it to select a subset of unlabeled data in the training process. In the last, we proposed a unified framework, called successive subspace learning [...]

By |September 15th, 2019|News|Comments Off on Congratulations to Yueru Chen for Passing Her PhD Defense|

MCL Research on Graph Embedding

Graph is a data representation model. Each data point is considered as a node and an edge/connection exists between nodes if there is any common characteristics. The relationship that exists between nodes is complex and attracts research in this domain. Several techniques have been developed like Deep Walk, Planetoid, Chebyshev, Graph Convolution Network, Graph Attention Network, Large Scale Graph Convolution Network, and so on, which focuse on exploring the behavior of the nodes based on their connectivity to different nodes. Graph models are often designed for tasks like Node classification, Edge/Link prediction, and has varied applications in social network, citation networks.

Currently we are developing a Graph Neural Network model for node classification task of a Graph. Feedforward based approach is adopted to learn the network parameters in a single forward pass using Graph Hop Method. The main idea is to learn the node’s representation making use of their hop’s (neighboring node’s) representation to better represent and learn from local to global attribute perspective through information exchange between the hops, by subsequently growing the dimension of the feature vector and reducing the dimensionality using SaaB transform.

Unlike the methods/techniques which are already developed, our model’s computational complexity is very low for the fact that no back propagation is utilized for learning the parameters of the network model, but through feedforward design the model learns in a single forward pass. The Graph Hop Method serves as a unique method for driving the model to train on very less training samples yet provide better accuracies/results for testing samples. Thus, the model is capable to train on very limited labelled data. Making use of only 5% of training samples, we are able to achieve the state of art performance [...]

By |September 8th, 2019|News|Comments Off on MCL Research on Graph Embedding|