Colloquia Announcements

Wednesday, 28th November 2018, 5:00pm, UB506

Constructing a robot's free space can be useful for fast and accurate navigation, but construction is difficult in high dimensions. In this talk I will present an algorithm for closely approximating the free space of a robot with 4 degrees of freedom (for example, a drone helicopter). Additionally, I will show how the free space can be utilized for safe and reliable navigation.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30 p.m. in the reception area of the 3rd floor of Ungar.

Wednesday, 14th November 2018, 5:00pm, UB506

There have been promising advances in brain machine interfaces. However, many challenges still remain before this technology can become practical. In this talk, I will discuss the main issues along with possible approaches to overcome them. In particular, I will introduce methods to translate neural signals to control external devices using reinforcement learning and a neural signal emulator to help accelerate brain machine interface research.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30 p.m. in the reception area of the 3rd floor of Ungar.

Wednesday, 7th November 2018, 5:00pm, UB506

The commercial remote sensing sector has experienced a period of rapid growth over the past decade. High-frequency, medium-resolution (3-5 meter GSD), electro-optical satellite imagery presents a unique and unprecedented opportunity to perform daily monitoring over wide areas. The 3GIMBALS team will discuss their PISCES project. PISCES is an automated maritime port activity monitoring application designed to unlock the potential of this new data source for intelligence analysts in the defense sector.  First, the team will outline the thought process and approach taken throughout the project including how they used agile software development, minimally viable products, and client-need discovery to move fast and delight the client. Next, they will walk the audience through the computer vision algorithm development, the types of time series analysis and anomaly detection algorithms applied on top of it, and the hardware-cloud-web hybrid infrastructure we use to deliver the service. Finally, they will discuss past and present challenges, as well as, share insights about potential future directions of the application into multi-source data fusion.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30 p.m. in the reception area of the 3rd floor of Ungar.

Wednesday, 31st October 2018, 5:00pm, UB506

An international race is currently underway to develop quantum computers and quantum algorithms. In this elementary introduction, we'll address such questions as: What is it? Why is it important? And where are we now?

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30 p.m. in the reception area of the 3rd floor of Ungar.

Wednesday, 24th October 2018, 5:00pm, UB506

The Open Source Enterprise (OSE) is under CIA's Directorate of Digital Innovation (DDI) and is the Agency's newest Directorate focused on accelerating innovation across the Agency's mission activities with cutting-edge digital and cyber tradecraft and IT infrastructure. Please join us to learn about the history, mission, and career opportunities in Computer Sciences, of this unique organization. Representatives from CIA and OSE will be available to answer your questions.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30 p.m. in the reception area of the 3rd floor of Ungar.

Wednesday, 17th October 2018, 5:00pm, UB506

Stock data is abundant and all around us.It is the goal of many organizations, groups and individuals to know the direction and movements of the overall stock market as well as for individual stocks. In this talk, I'll discuss how to calculate technical indicators from historical stock data, and how to create features and targets out of historical stock data. I will then discuss several machine learning algorithms which can be used to predict the future price of stocks. And finally, the resulting performance of these models will be evaluated and compared.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30 p.m. in the reception area of the 3rd floor of Ungar.

Wednesday, 10th October 2018, 5:00pm, UB506

Large-scale deep neural networks usually employ sophisticated internal signal propagation structures and parameter settings and thus the learning process and results are difficult to interpret and to interact with manual processes. We propose a novel pattern interpretation and interaction framework that combines the predictive accuracy of deep learning systems and the pattern/process transparency of the conventional statistical pattern recognition algorithms in the application context of content-based prediction of manual multimedia annotation labels. Our implementation of deep neural network based multimedia content analysis system employs convolutional neural network layers for modeling the video frame pixels and the audio spectrogram samples. Then the feature maps from different video segments are combined using recurrent neural networks for predicting the annotation labels. The proposed pattern interpretation and interaction framework feeds in the statistical signal patterns into the feature fusing and temporal modeling layers of the deep neural networks. Experiments show that the deep neural networks tuned to specific suboptimal configurations receive the highest amount of boost of the predictive analysis performance from the feature integration process of appending external multimedia signal patterns, while these signal patterns alone do not have meaningful predictive performance on the same predictive analysis tasks. The performance enhancement pattern of integrating the external statistical patterns also helps to establish the information dependency structures between the video/audio signal patterns to the predictive targets of manual annotation labels and forms cognitively relevant interpretations of the multimedia signal patterns.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30 p.m. in the reception area of the 3rd floor of Ungar.

Wednesday, 3rd October 2018, 5:00pm, UB506

In this seminar I will present my research on model-based biomedical image analysis. In order to extract information of objects from biomedical images (e.g. microscopy images of cells), mathematical models were built to represent the objects and describe their properties. Strategies and algorithms were developed to fit the object models to image data, and then information about the objects was obtained. I will show three applications including subcellular particle image analysis, cell image analysis, and aortic valve image analysis. I will focus on the models, strategies and algorithms in these applications. The advantage of model-based approaches is that the models are interpretable and could be physics-based, and the disadvantage is that it may be time-consuming to derive and use such models.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30 p.m. in the reception area of the 3rd floor of Ungar.

Wednesday, 26th September 2018, 5:00pm, UB506

After decades of languishing in university labs, machine learning has become respectable. Major companies have established entire divisions specializing on machine learning software, newspapers publish articles about spectacular applications, and the public is intrigued. A scientist who has spent a lifetime working in this field rejoices---and recapitulates. This is the time to pause, take a deep breath, and look back at the story of the discipline. What motivated its beginnings? Which major avenues has it followed? How has it conducted its business? Just as well, this is also a time to look ahead. Widely publicized success stories may have obscured some other benefits that machine learning can offer, beyond the now-so-popular recognition of complicated patterns. The talk will therefore briefly mention some unfairly neglected possibilities that may inspire future research.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30 p.m. in the reception area of the 3rd floor of Ungar.

Wednesday, 19th September 2018, 5:00pm, UB506

Hi-C data have been widely used to reconstruct chromosomal three-dimensional (3D) structures. One of the key limitations of Hi-C is the unclear relationship between spatial distance and the number of Hi-C contacts. Many methods used a fixed parameter when converting the number of Hi C contacts to wish distances. However, a single parameter cannot properly explain the relationships between wish distances and genomic distances. We addressed one of the key issues of using Hi-C data, that is, the unclear relationship between spatial distance and the number of Hi-C contacts, which was crucial to understanding significant biological functions, such as the enhancer-promoter interactions. We developed a new method to infer this converting parameter and pairwise Euclidean distances based on the topology of the Hi-C complex network. The inferred distances were modeled by clustering coefficients and multiple other types of constraints. We demonstrated the methodology on biological data and reconstructed a high-resolution 3D chromosomal structures of mouse male cells.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30 p.m. in the reception area of the 3rd floor of Ungar.

Wednesday, 12th September 2018, 5:00pm, UB506

Deep Convolutional Neural Networks have become the current state of the art in Computer Vision problems such as object and scene recognition. Most of their success has been attributed to the ability of these networks to learn a representation of the objects rather than using handcrafted features. Interestingly, these networks, which have been trained to perform high-level tasks, also display an intriguing ability to explain some aspects of visual neural processing in the brain. In our work, we exploit the representations learned by Deep Convolutional Networks to expand our understanding of what is currently known for primary visual cortex in the brain, to intermediate stages of the visual cortex. I will discuss how this work can have mutual impacts to inform models of the brain and to advance artificial intelligence.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30 p.m. in the reception area of the 3rd floor of Ungar.

Wednesday, 5th September 2018, 5:00pm, UB506

We deal with multi-objective optimization problems (MOOP) in our everyday lives. Some problems such as identifying the smallest subset of genes with the highest prediction accuracy of a genetic disease are simply not solvable in reasonable (polynomial) time and resort to metaheuristic approaches. Particle swarm optimization (PSO) is a simple, yet powerful metaheuristic evolutionary algorithm that has been successfully used in the last twenty years to solve complex MOOP. While powerful, such algorithms suffer from early and suboptimal convergence, leaving large areas of the search space unexplored. In this talk, I introduce COMB-PSO, a heuristic that solves such issues and further allows the minimization of the number of features and the enhancement of the convergence rate, precision, robustness and general performance.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30 p.m. in the reception area of the 3rd floor of Ungar.

Wednesday, 29th August 2018, 5:00pm, UB402

As of August 22nd 2018, total market capitalization of crypto currencies is about $208 trillion. Out of that Bitcoin's share is about $111.2 trillion. This talk introduces blockchain, the technology behind cryptocurrencies. Satoshi Nakamoto proposed Blockchain in 2008 to serve as the public transaction ledger of the cryptocurrency bitcoin. Blockchains that are readable by the public is widely used for cryptocurrencies. A block in a blockchain contains cryptographic hash of the previous block, a timestamp, and data. We will discuss algorithms for `mining' blocks. We will also discuss other potential applications of blockchain.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30 p.m. in the reception area of the 3rd floor of Ungar.

Previous Colloquia Announcements