Organizational Unit:

School of Computer Science

Permanent Link

https://hdl.handle.net/1853/70781

Parent Organization

Organizational Unit

College of Computing

ArchiveSpace Name Record

https://finding-aids.library.gatech.edu/agents/corporate_entities/945

Full item page

Publication Search Results

Now showing 1 - 10 of 457

How to Approach Right, the Regulation of Educational AI?

(Georgia Institute of Technology, 2019-12-24) Samovich, Valery

The purpose of this research is to find the right approach to regulate educational AI. First, I analyzed the existing AI initiatives, AI regulatory approaches and identified the core elements of AI regulations. Second, I conducted a survey and find out what values are important for regular students. And, finally, I summarized and formulated the approach for the regulation of Educational AI.
Real time detection of traffic signs on mobile device

(Georgia Institute of Technology, 2019-12-09) Six, Nicolas

In this work we propose a new approach to the object detection problem using Deep Neural Network, in the context of traffic sign detection. Our approach simplifies the detection head complexity by making the requirement for localization lower and taking advantage of our particular task to make the feature extraction model smaller. This strategy allows to create a model running at 88 frames per second on a four years old smartphone, a Samsung S6 (SM-G920T), while maintaining a mAP@50 at 55% and mAP@25 at 68%. To get these results, we created a way to generate data for training based on random geometrical shapes that allows to initialize the weights of our model before training on real data. To the best of our knowledge this model provides the best accuracy over speed ratio for the detection of traffic signs on mobile device at the moment.
Next Generation Learning Platform - Reference Architecture Based on Open Standard

(Georgia Institute of Technology, 2019-12-07) Kelklie, Moges

There are hundreds of companies developing learning tools and capabilities; however, there are not many papers published on how these technologies are interconnected to provide a complete learning architecture. Because of the lack of comprehensive open learning architecture, education companies are forced to piece together many technologies and hardwire them through a non-standard integration. In recognizing the lack of progress on learning management tools, Educause proposed a conceptual framework called the next-generation digital learning environment (NGDLE). This paper explores NDGLE and suggests a reference architecture based on open standards.
Semantic Mapping and Reasoning

(Georgia Institute of Technology, 2019-12) Chen, Kevin Julian

Rich, yet efficient knowledge processing is one of the key problems in modern autonomous robotics. The Robot Autonomy and Interactive Learning (RAIL) Lab at the Georgia Institute of Technology has developed a new knowledge processing framework named Robot Common Sense Embedding (RoboCSE), which leverages multi-relational embeddings to learn object affordances, locations, and materials. This project aims to test the capabilities of RoboCSE for household robots by building a perception pipeline, which outputs a semantic map (i.e. map with object labels). The perception pipeline consists of two main components: a Simultaneous Localization and Mapping (SLAM) algorithm to build an occupancy map and two object classifiers to label objects in the map. We hope to integrate the semantic map into RoboCSE to test RoboCSE’s ability to perform high-level task planning and knowledge sharing.
On Formula Embeddings in Neural-Guided SAT Solving

(Georgia Institute of Technology, 2019-12) Dumenci, Mert

Branching heuristics determine the performance of search-based SAT solvers. We note that recently, Neural Machine Learning approaches have been proposed to learn such heuristics from data. The first step in learning a branching heuristic is a transformation from the space of Boolean formulas to a vector space. We note that there is no canonical transformation: techniques such as message-passing networks and LSTMs have been proposed to embed formulas into R^n. We build a novel dataset of an approximate optimal heuristic and compare the estimation performance of models with different embedding methods. We show that for performant models, embedding methods need to represent the structural invariances of Boolean formulas: similar to CNNs and spatially local data such as images.
Extracting information from gameplay videos using machine learning techniques and its varieties

(Georgia Institute of Technology, 2019-12) Luo, Zijin

The ability to extract sequences of game events for high-resolution e-sport games has traditionally required access to the game’s engine. This difficulty serves as a barrier to groups who don’t possess this access. It is possible to apply deep learning to derive these logs from gameplay video, but it requires computational power that serves as an additional barrier. These groups would benefit from access to these logs, such as small e-sport tournament organizers who could better visualize gameplay to inform both audience and commentators. In this dissertation, we present a combined solution to reduce the required computational resources and time to apply a convolutional neural network (CNN) to extract events from e-sport gameplay videos. This solution consists of techniques to train CNN faster and methods to execute predictions more quickly. These techniques expand the types of machines capable of training and running these models, which in turn extends access to extracting game logs with this approach. We evaluate the approaches in the domain of DOTA2, one of the most popular competitive e-sport games. Our results demonstrate our approach outperforms standard backpropagation baselines.
Stable Model Predictive Path Integral Control for Aggressive Autonomous Driving

(Georgia Institute of Technology, 2019-12) Capuano, Matthieu Jean Baptiste Jean-Baptiste

A common challenge with sampling based Model Predictive Control (MPC) algorithms operating in stochastic environments is ensuring stable behavior under sudden state disturbances. Model Predictive Path Integral (MPPI) control is an MPC algorithm that can optimize control of non-linear systems subject to non-differentiable cost criteria. It iteratively computes optimal control sequences by re-using the sequence optimized at the previous timestep as a warm start for the current iteration, which allows rapid convergence thus making it real time capable. This approach is successful in producing a diverse set of behaviors, the most impressive being its ability to control systems at the limits of handling. However, a strong unexpected state disturbance can make the previous control sequence an unsafe initialization for the new state and can result in undesired behavior. In this work, we address this problem by implementing a path tracker that produces control sequences that are used as the initializers for the current timestep, instead of simply re-using the sequence from the previous timestep. The path tracker iteratively computes control sequences that can guide the system to low-cost regions and feeds them into the MPPI framework as a sampling reference. This enforces the algorithm to sample behaviors normally distributed around controls that guide the state back to low-cost regions, even in cases where the state drastically changes. The additional advantage of our method is that it retains the ability to sample diverse and dynamically feasible controls, thus maintaining its ability for motion at the limits of handling. We experimentally verify this method on the AutoRally autonomous research platform, a one-fifth scale race car for aggressive driving tasks, and compare its performance against the most recently published results of MPPI for autonomous driving.
Robot Calligraphy using Pseudospectral Optimal Control and a Simulated Brush Model

(Georgia Institute of Technology, 2019-12) Chen, Jiaqi

Chinese calligraphy is unique and has great artistic value but is difficult to master. In this paper, we make robots write calligraphy. Learning methods could teach robots to write, but may not be able to generalize to new characters. As such, we formulate the calligraphy writing problem as a trajectory optimization problem, and propose a new virtual brush model for simulating the dynamic writing process.Our optimization approach is taken from pseudospectral optimal control, where the proposed dynamic virtual brush model plays a key role in formulating the objective function to be optimized. We also propose a stroke-level optimization to achieve better performance compared to the character-level optimization proposed in previous work. Our methodology shows good performance in drawing aesthetically pleasing characters.
Stranger Danger: Educational Game for Cybersecurity Awareness

(Georgia Institute of Technology, 2019-12) Ren, Ziang

Cybersecurity is a concern for both organizations and individual users. Although there are a variety of security tools available, the number of cybersecurity incidents is still high. One cause of this phenomenon is that users are typically unaware or underestimating the potential negative consequences of their insecure behaviors on the internet. This leads to a lack of motivation among users for adopting security tools that require additional efforts in order to access a desired service. Conventional methods in promoting people s awareness of cybersecurity such as information sessions hosted by cybersecurity professionals have been shown to be ineffective. Video games is a novel approach in cybersecurity education has achieved some success in training cybersecurity professionals. However, whether video games are an effective method for educating the general public about the basic concepts of cybersecurity remains untested. This study presents Stranger Danger, a video game that simulates real-world exploits and teaches its users via negative reinforcements. The purpose of the study is to examine whether video games are effective in teaching the general public about basic cybersecurity concepts.
Evaluating Off-Center Head-Worn Display

(Georgia Institute of Technology, 2019-12) Ramakrishnan, Rohan

Several studies have highlighted the advantages of using mobile augmented reality systems to assist with various tasks over traditional paper-based methods. However, these interfaces are often located in users’ primary field of view which causes interference with users’ vision and presents several disruptions. In this paper, a new ”off-center” display type is prototyped and compared across other displays using a coloring task. Metrics such as completion time, errors, and workload are collected and used to find tradeoffs between different display types and determine their feasibility.