Title: Undersampled MRI Reconstruction, Uncertainty Measurement, and k-space Acquisition Planning
Speaker: Zizhao Zhang, Ph.D student - University of Florida
Time: 4:30 pm to 5:30 pm
Date: September 27, 2018
Location: Room E301J CSE Bldg
Abstract: Magnetic resonance imaging (MRI) is the standard for picturing anatomy in radiology. MRI is saver compared with computed tomography (CT) imaging because MRI has no ionizing radiation side effects. However, the problem of MRI is its time-consuming scanning process. Consequently, this problem causes uncomfortable exam experience and high cost, making it less accessible to patients. Accelerating MRI is a critical medical imaging problem in healthcare. Its successful application has the potential to substantially improve the current situation of MRI, and allow MRI to be accessible to a larger population eventually. Facebook AI Research recently launched a collaborative heathcare project (see news at CNN, Forbes, and MIT technology review). The goal is to build an open research platform to speedup its development using artificial intelligence (AI). In this talk, Zizhao will present a novel deep learning algorithm he developed during his internship. The algorithm is about uncertainty measurement and active k-space acquisition planning in MRI reconstruction. The proposed algorithm offers a level of interpretability of black-boxed neural networks. And it is applicable onto a MRI machine to guide its signal acquisition and thereby maximize the MRI acceleration factor and reconstruction effectiveness and efficiency. Moreover, the proposed algorithm not only shows high-quality reconstruction results but also paves the way towards to more clinically applicable MRI reconstruction.
Title: DensePose: Dense Human Pose Estimation In The Wild
Speaker: Siegel,Scott N, Ph.D student - University of Florida
Time: 4:30 pm to 5:30 pm
Date: September 11, 2018
Location: Room L120 Darwin, Marston Science Library Bldg
Abstract: Deep Reinforcement Learning has received a lot of attention due to Google DeepMind's successes in Atari and Go, and OpenAI's recent success at Dota 2. In essence, reinforcement learning is a framework for training intelligent agents to solve complex tasks by interacting with their environment. Fundamental to reinforcement learning is the concept of decision-making under uncertainty. Notably, deep neural networks have enabled many of the classic reinforcement learning algorithms to be applied to more complex and high-dimensional domains. In this talk, an overview of the major milestones in deep reinforcement learning will be presented, along with a brief introduction to various state-of-the-art algorithms.
Title: Densely Labeling Large-Scale Satellite Images with Generative Adversarial Networks
Speaker: Xiaohui Huang, Ph.D student - University of Florida
Time: 4:30 pm to 5:30 pm
Date: August 8, 2018
Location: Room E440 CSE Bldg
Abstract: Building an efficient and accurate pixel-level labeling framework for large-scale and high-resolution satellite imagery is an important machine learning application in the remote sensing area. Due to the very limited amount of the ground- truth data, we employ a well-performing superpixel tessellation approach to segment the image into homogeneous regions and then use these irregular-shaped regions as the foundation for the dense labeling work. A deep model based on generative adversarial networks is trained to learn the discriminating features from the image data without requiring any additional labeled information. In the subsequent classification step, we adopt the discriminator of this unsupervised model as a feature extractor and train a fast and robust support vector machine to assign the pixel-level labels. In the experiments, we evaluate our framework in terms of the pixel-level classification accuracy on satellite imagery with different geographical types. The results show that our dense-labeling framework is very competitive compared to the state-of-the-art methods that heavily rely on prior knowledge or other large-scale annotated datasets.
Title: Visual Object Tracking: An overview
Speaker: Pan He, Ph.D student - University of Florida
Time: 4:30 pm to 5:30 pm
Date: July 25, 2018
Location: Room E440 CSE Bldg
Object tracking in realistic scenarios is a difficult problem, therefore, it remains a most active area of research in computer vision. In this talk, I will give one overview on visual tracking problem and indicate future trends.
Title: Deep Reinforcement Learning: An Overview
Speaker: Patrick Emami, Ph.D student - University of Florida
Time: 4:30 pm to 5:30 pm
Date: July 11, 2018
Location: Room E440 CSE Bldg
Abstract: Deep Reinforcement Learning has received a lot of attention due to Google DeepMind's successes in Atari and Go, and OpenAI's recent success at Dota 2. In essence, reinforcement learning is a framework for training intelligent agents to solve complex tasks by interacting with their environment. Fundamental to reinforcement learning is the concept of decision-making under uncertainty. Notably, deep neural networks have enabled many of the classic reinforcement learning algorithms to be applied to more complex and high-dimensional domains. In this talk, an overview of the major milestones in deep reinforcement learning will be presented, along with a brief introduction to various state-of-the-art algorithms.
Title: Multi-Agent Reinforcement Learning for Integrated Network of Adaptive Traffic Signal Controllers (MARLIN-ATSC)
Speaker: Rahul Sengupta, Ph.D student - University of Florida
Time: 4:30 pm to 5:30 pm
Date: June 26, 2018
Location: Room E440 CSE Bldg
Abstract: Population is steadily increasing worldwide, resulting in intractable traffic congestion in dense urban areas. Adaptive traffic signal control (ATSC) has shown strong potential to effectively alleviate urban traffic congestion by adjusting signal timing plans in real time in response to traffic fluctuations to achieve desirable objectives (e.g., minimize delay). Efficient and robust ATSC can be designed using a multiagent reinforcement learning (MARL) approach in which each controller (agent) is responsible for the control of traffic lights around a single traffic junction. Applying MARL approaches to the ATSC problem is associated with a few challenges as agents typically react to changes in the environment at the individual level, but the overall behavior of all agents may not be optimal. This paper presents the development and evaluation of a novel system of multiagent reinforcement learning for integrated network of adaptive traffic signal controllers (MARLIN-ATSC). MARLIN-ATSC offers two possible modes: 1) independent mode, where each intersection controller works independently of other agents; and 2) integrated mode, where each controller coordinates signal control actions with neighboring intersections. MARLIN-ATSC is tested on a large-scale simulated network of 59 intersections in the lower downtown core of the City of Toronto, ON, Canada, for the morning rush hour. The results show unprecedented reduction in the average intersection delay ranging from 27% in mode 1 to 39% in mode 2 at the network level and travel-time savings of 15% in mode 1 and 26% in mode 2, along the busiest routes in Downtown Toronto.
Title: Adaptive Adversarial Attack on Scene Text Recognition
Speaker: Xiaoyong, Yuan, Ph.D student - University of Florida
Time: 4:30 pm to 5:30 pm
Date: June 19, 2018
Location: Room E440 CSE Bldg
Abstract: Recent studies have shown that state-of-the-art deep learning models are vulnerable to the inputs with small perturbations (adversarial examples). We observe two critical obstacles in adversarial examples: (i) Strong adversarial attacks require manually tuning hyperparameters, which take longer time to construct a single adversarial example, making it impractical to attack real-time systems; (ii) Most of the studies focus on non-sequential tasks, such as image classification and object detection. Only a few consider sequential tasks. Despite extensive research studies, the cause of adversarial examples remains an open problem, especially on sequential tasks. We propose an adaptive adversarial attack, called AdaptiveAttack, to speed up the process of generating adversarial examples. To validate its effectiveness, we leverage the scene text detection task as a case study of sequential adversarial examples. We further visualize the generated adversarial examples to analyze the cause of sequential adversarial examples. AdaptiveAttack achieved over 99.9% success rate with 3 ∼ 6× speedup compared to state-of-the-art adversarial attacks.
Title: Introduction to traffic engineering and recent developments
Speaker: Mahajan, Dhruv, Ph.D student - University of Florida
Time: 4:30 pm to 5:30 pm
Date: June 13, 2018
Location: Room E440 CSE Bldg
Abstract: In the field of traffic engineering, engineers must look at the whole picture in order to maximize traffic flow and reduce instanced of congestion: the movement of vehicles on roads and highways; the movement of pedestrians. Common objectives of traffic engineering included the following: i) Providing high efficient traffic flow through ample research and innovative design efforts. ii) To produce free flow of traffic. iii) Use research to design roadways and highways that increase traffic safety (strategic implementation of stop signs, traffic signs, and traffic lights) The talk will also cover the upcoming changes to the filed including the use of sensors, networked and pervasive (IOT) devices and their possible impact.
Title: Intro to Pytorch
Speaker: Caleb Bryant, Undergraduate student - University of Florida
Time: 4:30 pm to 5:30 pm
Date: May 30, 2018
Location: Room E440 CSE Bldg
Abstract: As neural networks grow larger and more complex, new tools are needed to manage these networks. Pytorch is a popular Deep Learning Framework that has grown popular with researchers in recent years, and it can be used to both simplify network design and take advantage of modern computing power. In this talk, we briefly examine some of the design considerations involved with building Pytorch, explore the functional unit of Pytorch–the Tensor, and look at how to apply the power of Pytorch to a Kaggle image classification problem.
Title: Dynamic Load Balancing for Compressible Multiphase Turbulence
Speaker: Keke Zhai, PhD student - University of Florida
Time: 4:30 pm to 5:30 pm
Date: May 23, 2018
Location: Room E440 CSE Bldg
Abstract: CMT-nek is a new scientific application for performing high fidelity predictive simulations of particle laden explosively dispersed turbulent flows. CMT-nek involves detailed simulations, is compute intensive and is targeted to be deployed on exascale platforms. The moving particles are the main source of load imbalance as the application is executed on parallel processors. In a demonstration problem, all the particles are initially in a closed container until a detonation occurs and the particles move apart. If all processors get an equal share of the fluid domain, then only some of the processors get sections of the domain that are initially laden with particles, leading to disparate load on the processors. In order to eliminate load imbalance in different processors and to speedup the makespan, we present different load balancing algorithms for CMT-nek on large scale multi-core platforms consisting of hundred of thousands of cores. The detailed process of the load balancing algorithms are presented. The performance of the different load balancing algorithms are compared and the associated overheads are analyzed. Evaluations on the application with and without load balancing are conducted and these show that with load balancing, simulation time becomes faster by a factor of up to 9.97.
Title: Im2Flow: Motion Hallucination from Static Images for Action Recognition
Speaker: Xiaohui Huang, PhD student - University of Florida
Time: 4:30 pm to 5:30 pm
Date: May 16, 2018
Location: Room E440 CSE Bldg
Abstract: Existing methods to recognize actions in static images take the images at their face value, learning the appearances—objects, scenes, and body poses—that distinguish each action class. However, such models are deprived of the rich dynamic structure and motions that also define human activity. We propose an approach that hallucinates the unobserved future motion implied by a single snapshot to help static-image action recognition. The key idea is to learn a prior over short-term dynamics from thousands of unlabeled videos, infer the anticipated optical flow on novel static images, and then train discriminative models that exploit both streams of information. Our main contributions are twofold. First, we devise an encoder-decoder convolutional neural network and a novel optical flow encoding that can translate a static image into an accurate flow map. Second, we show the power of hallucinated flow for recognition, successfully transferring the learned motion into a standard two-stream network for activity recognition. On seven datasets, we demonstrate the power of the approach. It not only achieves state-of-the-art accuracy for dense optical flow prediction, but also consistently enhances recognition of actions and dynamic scenes.
Title: Sinkhorn Policy Gradient for Learning with Permutations
Speaker: Patrick Emami, PhD student - University of Florida
Time: 4:30 pm to 5:30 pm
Date: May 9, 2018
Location: Room E440 CSE Bldg
Abstract: A large number of problems at the intersection of combinatorics and computer science require finding a permutation that optimally matches, ranks, or sorts some data. Data-driven algorithms can potentially use non-differentiable task-specific objectives as a reward function to augment the learning process. In this paper, we propose the Sinkhorn Policy Gradient (SPG) algorithm for learning policies directly on the set of N x N permutation matrices. SPG uniquely decouples representation learning of problem instances from the highly-structured action space of permutations by leveraging the recently introduced Sinkhorn layer, making SPG applicable to many problem domains. Our empirical results show that SPG can perform competitively on sorting, the Euclidean TSP, and planar maximum weight matching.
