News

We are so happy to welcome our new MCL member, Pranav Kadam! Here is a short interview with Pranav.

1. Could you briefly introduce yourself and your research interests?
My name is Pranav Kadam and am a graduate student at USC pursuing Master’s degree in Electrical Engineering. I grew up in a city called Pune which is in the western part of India. Prior to joining USC in Fall 2018, I completed my Bachelor’s degree in Electronics and Telecommunications Engineering from Savitribai Phule Pune University. In my undergraduate coursework, I found topics like image processing and neural networks quite interesting. I did some projects in these fields and read a lot a regarding recent developments then, such as CNNs. I am continuing my research and specialization in Computer Vision and Machine Learning at USC.

2. What is your impression about MCL and USC?
USC is a great urban campus with excellent educational and research facilities. A lot of focus is given on mastering the fundamental topics and concepts for a long-term career in the field of interest. In addition, Los Angeles is a large metropolis and there are plenty of options for entertainment, food and sightseeing, all very nearby campus, making it an ideal city to be in during graduate studies. MCL is a team of talented and passionate students headed by Prof. Kuo who is very experienced, insightful and supportive. Lab activities like research meetings and weekly seminars ensure that everyone is aware of each other’s work and there is a good bonding among members.

3. What is your future expectation and plan in MCL?
I would like to work under the guidance of Prof. Kuo and develop the necessary skills to be a successful researcher. [...]

We are so glad to welcome our new MCL member, Yifan Wang! Here is a short interview with Yifan:

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

My name is Yifan Wang, a graduate student in Department of Electrical and Engineering of USC. I received my bachelor’s degree in electrical engineering from Fudan University. When I was an undergraduate student, I learnt courses from multi-fields including digital and analog circuit, computer architecture etc. Among these, I found my interests lies in vision fields which is of vital importance to both human and machine.

2. What is your impression about MCL and USC?

I have been USC for two semesters. It is a really nice place for studying, since there are few entertainment sites outside school along with the super good sun shine prohibit me from going outside regularly. MCL is a large warn family where I met lots of talent people. It would be nice to make new friends and learn new things in MCL.

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

I would like to learn more theories and mathematics related to computer vision which would be the foundation of my future research. Besides, I would like to gain more practice in deep learning fields which are highly hardware relied.

With the rise of visualization, animation and autonomous driving applications, the demand for 3D point cloud analysis and understanding has rapidly increased. Point Cloud is a kind of data obtained from lidar scanning which contains abundant 3D information. ModelNet40 is a point cloud dataset contains 40 classes of objects. In this project, we use ModelNet40 dataset for the analysis and evaluation of point cloud classification. Many of the recent works focus on developing end to end algorithm like other convolutional neural networks for images. However, object and scene understanding with Convolutional Neural Networks (CNNs) on 3D volumetric data is still limited due to its high memory requirement and computational cost. For some simple tasks like classification, this method is too much.

An interpretable CNN design based on the feedforward (FF) methodology [1] without any backpropagation (BP) was recently proposed by the Media Communications Lab at USC. The classification baseline is composed by four Saab units, each unit contains KNN query, space grouping and Saab transform, and between units we use farthest sampling to improve efficiency. We are still working on it to improve as much as possible. Our goal is to catch up with the state-of-the-art results and show that FF design is powerful and useful. The advantages of the FF design methodology are multiple folds. It is completely interpretable. It demands much less training complexity and training data. Furthermore, it can be generalized to weakly supervised or unsupervised learning scenarios in a straightforward manner. The latter is extremely important in real world application scenarios since data labeling is very tedious and expensive.

The advantages of the FF design methodology are multiple folds. It is completely interpretable. It demands much less training complexity and training data. [...]

Congratulations to Harry Yang for passing his defense on May 7, 2019! Let us hear what he would like to say about his defense and an abstract of his thesis.

“In the thesis, we tackle the problem of translating faces and bodies between different identities without paired training data: we cannot directly train a translation module using supervised signals in this case. Instead, we propose to train a conditional variational auto-encoder (CVAE) to disentangle different latent factors such as identity and expressions. In order to achieve effective disentanglement, we further use multi-view information such as keypoints and facial landmarks to train multiple CVAEs. By relying on these simplified representations of the data we are using a more easily disentangled representation to guide the disentanglement of image itself. Experiments demonstrate the effectiveness of our method in multiple face and body datasets. We also show that our model is a more robust image classifier and adversarial example detector comparing with traditional multi-class neural networks.

“To address the issue of scaling to new identities and also generate better-quality results, we further propose an alternative approach that uses self-supervised learning based on StyleGAN to factorize out different attributes of face images, such as hair color, facial expressions, skin color, and others. Using pre-trained StyleGAN combined with iterative style inference we can easily manipulate the facial expressions or combine the facial expressions of any two people, without the need of training a specific new model for each of the identity involved. This is one of the first scalable and high-quality approach for generating DeepFake data, which serves as a critical first step to learn a more robust and general classifier against adversarial examples.”

Harry also shared about his Ph.D. experience:

“Firstly, I would [...]

Our long-term goal is to build intelligent systems that can perceive their visual environment and understand the linguistic information, and further make an accurate translation inference to another language. However, most multi-modal translation algorithms are not significantly better than an off-the-shelf text-only machine translation (MT) model. There remains an open question about how translation models should take advantage of visual context, because from the perspective of information theory, the mutual information of two random variables I(X; Y) will always be no greater than I(X, Z; Y) where Z is the additional visual input. This conclusion makes us believe that the visual content will hopefully help the translation systems.

Since the standard paradigm of multi-modal translation always considers the problem as a supervised learning task, the parallel corpus is usually sufficient to train a good translation model, and the gain from the extra image input is very limited. We however argue that the text-only UMT is fundamentally an ill-posed problem, since there are potentially many ways to associate target with source sentences. Intuitively, since the visual content and language are closely related, the image can play the role of a pivot “language” to bridge the two languages without paralleled corpus, making the problem “more well-defined” by reducing the problem to supervised learning.

We tackle the unsupervised translation with a multi-modal framework which includes two sequence-to-sequence encoder-decoder models and one shared image feature extractor in order to achieve the unsupervised translation. We employ transformer in both the text encoder and decoder of our model and design a novel joint attention mechanism to simulate the relationships among the language and visual domains.

Succinctly, our contributions are three-fold:

We formulate the multi-modal MT problem as unsupervised setting that fits the real [...]