URL: /topics/ai-ml/computer-vision.html
PATH: _topics/ai-ml/computer-vision.md
Computer Vision
Hands-free tracking of chemical substances in research lab
Scientists can register their lab samples by holding up bottle to the camera. Captured images of the bottle labels are pre-processed locally, then processed for OCR and registration in the cloud. Use custom CNN, Amazon Textract, Amazon Rekognition.
Serial number recognition
Read serial numbers (SNs) from machine parts. Part of the SN is cast in mold, part is engraved as dot peen. Use pre-trained DNN models for object detection and object recognition. Post processing corrects SNs based on format rules to 100% accuracy. Use augmentation of labeled images to increase training set 100 fold.
Item counting on shopfloor
Quality control to validate shipping label with stack of materials before loading on the truck. OCR to read shipping label, custom image transformations to count number of items in stack, determine dimensions of materials.
Defect detection in manufacturing
Qaulity control for lamination process of sheet metal. Image classification to detect work material. Unsupervised annomaly detection of defects.
Multi-lane vehicle tracking for high-way traffic flow modeling
Data collection along 5-lane Inter-state highway to measure vehicle speed, spacing between vehicles, accelleration, and lane change frequency. Use commercially available technology for traffic control. Data is used to calibrate traffic flow models that predict travel time and risk of congestions.
Crowd tracking at public event to detect disturbances
Data on pedestrian movement at public gatherings was collected using commercially available tracking devices, which was subsequently utilized to calibrate the social force model.
Image categorization of vendor-uploaded photos on e-commerce site for improved search experience
The e-commerce site introduced a new comparative view of listed products categorized by their images. Because many images that were provided by vendors are not properly categorized, about 1 million images needed to be processed before launch. Use pre-trained CNN for image classification. Add extra layers to align with application specific classes.
Fraud detection on delivery site for raw materials using surveillance video
Delivery trucks are weighted upon entrance and exit of the site. Use surveillance video at weight stations to verify proper position of track on scale, type raw material, and that same truck enters and exists under the shipping number. Use custom feature generation, classification models, and image similarity.
Target recognition on hyper-spectral infrared imagery
Object detection in hyper-spectral infrared images with over 500 channels. Object size ranges from sub-pixel to few pixels. Develop image quality measure to compare performance of automated target recognition systems. Synthesize image data with multi-spectral ray tracer.
URL: /topics/ai-ml/demand-forecasting.html
PATH: _topics/ai-ml/demand-forecasting.md
Demand Forecasting, Time-series Analysis
Demand forecasting for launch of new consumer product
Forecast demand of new product prior to launch. Custom DNN time-series model trained on similar products.
Weekly demand forecasting for retail merchandise with wide ranging velocities
Demand forecasting for weekly re-ordering of about 60,000 products. Use ensemble of multiple models with product level selection on previous weeks.
Cost optimization of shipping materials leveraging time-series forecasting
Use probabilistic demand forecasting. Model cost of inventory and increased spending when using larger packaging.
Understand temporal patterns on e-commerce sites
Time series modeling for seasonal patterns: daily and weekly patterns of site visits, seasonal trends and impact of holidays. Seasonal demand of products.
Clinical Trials
Ensure meeting timelines for licensing and distribution in clinical trials leveraging time-series modeling and supply-chain simulation.
URL: /topics/ai-ml/indoor-localization.html
PATH: _topics/ai-ml/indoor-localization.md
Indoor Localization
Collect customer data in brick-and-mortar retail spaces: duration of stay, return visits, zones. Develop IoT devices for smart-phone tracking, laboratory and field study, evaluate accuracy, feasibility, and privacy.
URL: /topics/ai-ml/inventory-estimation.html
PATH: _topics/ai-ml/inventory-estimation.md
Inventory Estimation
Counting inventory of non-countable products, such as materials cut from roles or spools, or taken from containers or tanks, can be time-consuming and disruptive to operations if wall-to-wall counting is conducted.
At the cut-order fulfillment center, we addressed this challenge by developing a machine learning model based on a probabilistic hidden Markov chain. This model, trained using reinforcement learning with an evolutionary algorithm, significantly reduced inventory errors by over 50%.
URL: /topics/ai-ml/knowledge-graphs.html
PATH: _topics/ai-ml/knowledge-graphs.md
Knowledge Graphs
Cross-organizational data index
Build comprehensive data index across functional domains: research, manufacturing, distribution, marketing etc. Implement index as labeled property graph on graph database. Design collaborative tool for subject matter experts to define ontology and mapping of data sources to graph entities. Implement ETL workflow to update data index.
Improved document search
Use entity recognition on text documents to build knowledge graph. Combine semantic queries with graph search.
Rule-based learning and reasoning
Map knowledge graph to predictate logic. Use inductive logic programming to learn rules from examples.
URL: /topics/ai-ml/natural-language-processing.html
PATH: _topics/ai-ml/natural-language-processing.md
Natural Language Processing & Understanding
Build search index for electronic lab notebooks with attached documents
Scientists need to search through electronic lab notebooks and their attachments to retrieve data points from experiments. Use Amazon Textract and NLP processes to extract relevant data points from documents, use OpenSearch as searchable database.
Key-term extraction from contracts and addendums
Legal staff manually reviewed contract with business partners and addendums to keep an up-to-date database of the terms. Employ natural language processing (NLP) techniques to extract key terms, rule-based logic to update database with extracted key-terms from addendums.
Data extraction from news releases, social media posts, and emails for financial advisory firm
To keep an up-to-date database on mergers and acquisitions (M&A) deals, their terms, and their status, financial advisors rely on information from various sources such as news releases, social media posts, and emails. Employ natural language processing (NLP) techniques to efficiently process this information and disambiguate the relevant data.
Identify parties, their roles, relationships and actions from court filings
Build knowledge graph from narratives
Building situational knowledge graph from data
Classify and extract form data from health-care claims
Reduce response time for processing health-care claim by improved routing to specialist and automated extraction of key terms. Use Amazon Textract for OCR and key term extraction. Custom rules for post-processing and triaging.
Tagging of advertisements for improved marketing analysis
Advertiser was limited in their marketing analysis due to incomlete and inaccurate tagging of ads. Use NLP multi-class labeling to automatically tag advertisement artifacts.
URL: /topics/ai-ml/product-quality.html
PATH: _topics/ai-ml/product-quality.md
Product Quality, Manufacturing Processing Control
Investigate deterioration of product quality leveraging data from control systems, equipment sensors, raw materials. Integrate data lake for pro-active analytics and visualization. Predict product quality based on process data. Prioritize alarm messages for operators leveraging machine learning models. Mobile app for operators with interactive dashboard and manual.
URL: /topics/ai-ml/recommender-systems.html
PATH: _topics/ai-ml/recommender-systems.md
Recommender Systems
Email campaign with personalized recommendations based on customers viewing history
Collborative filtering based on page views. Multi-week campaign with customized recommendations.
Product recommendations on e-commerce site leveraging visual similarities
Product-similarity based recommendation based on visual features, including geometrical shapes, color distribution, and texture.
URL: /topics/sharing/lectures-and-workshops.html
PATH: _topics/sharing/lectures-and-workshops.md
Lectures and Workshops
Lectures
Course Number
Title
Level
Institution
CCIS 105
Programming Principles I
undergraduate
CAU
CCIS 106
Programming Principles II
undergraduate
CAU
CCIS 121
Introduction to Computer Systems
undergraduate
CAU
CCIS 223
Data Structures
undergraduate
CAU
CCIS 227
Deductive Systems
undergraduate
CAU
CCIS 371
Computer Algorithms
undergraduate
CAU
CCIS 374
Database Systems
undergraduate
CAU
CCIS 375
Artificial Intelligence
undergraduate
CAU
CCIS 413
Introduction to Robotics
undergraduate
CAU
CCIS 416
Introduction to High Performance Computing
undergraduate
CAU
CCIS 671
Algorithm Design and Analysis
graduate
CAU
CCIS 672
Computer Organization
graduate
CAU
CCIS 673
Operating Systems Design
graduate
CAU
CCIS 674
Database Design
graduate
CAU
CCIS 675
Artificial Intelligence
graduate
CAU
CCIS 676
Theory of Programming Languages Design
graduate
CAU
CCIS 683
Algorithms for Parallel Computers
graduate
CAU
CCIS 711
Image Processing
graduate
CAU
MSDA 8010
Data Programming
graduate
GSU
MSDA 8150
Machine Learning for Analytics
graduate
GSU
MSDA 8650
Advanced Deep Learning with Business Applications
graduate
GSU
Workshops
Title
Venue
Audience
Optimization Workshop
GSU
MS Data and Analytics students
AWS for Data Scientists
GSU
MS Data and Analytics students
Best Programming Practices for Data Scientists and Analysts
GSU
Professionals
URL: /topics/simulation/adaptive-sampling-by-histogram-equalization.html
PATH: _topics/simulation/adaptive-sampling-by-histogram-equalization.md
Adaptive Sampling by Histogram Equalization
Development of an adaptive sampling method that efficiently varies the sampling rate in local regions of a function based on the distribution of already collected samples. The algorithm does not rely on gradients in the parameter space and therefore allows to create accurate representation with less collected or computed sample points. In cases where he acquisition of samples is expensive, like computer simulations or experiments, ASHE has the advantage of requiring significantly less data points.
URL: /topics/simulation/air-corridors.html
PATH: _topics/simulation/air-corridors.md
Air Corridors
Air corridors are an integral part of the advanced air mobility infrastructure.
They are the virtual highways in the sky for transportation of people and cargo in the controlled airspace at an altitude of around 1000 ft. to 2000 ft. above the ground level.
This paper presents fundamental insights into the design of air corridors with high operational efficiency as well as zero collisions. It begins with the definitions of air cube, skylane or track, intersection, vertiport, gate, and air corridor. Then, a multi-layered air corridor model is proposed. Traffic at intersections is analyzed in detail with examples of vehicles turning in different directions.
The concept of capacity of an air corridor is introduced along with the nature of distribution of locations of vehicles in the air corridor and collision probability inside the corridor are discussed. Finally, the results of simulations of traffic flows are presented.
URL: /topics/simulation/dynamics-of-pedestrian-crowds.html
PATH: _topics/simulation/dynamics-of-pedestrian-crowds.md
Dynamics of Pedestrian Crowds
Development of a microscopic simulation model based on the concept of social forces. Study of the emergence of self-organization phenomena in crowds, and formation of trails.
URL: /topics/simulation/highway-traffic.html
PATH: _topics/simulation/highway-traffic.md
Vehicular Traffic
Predict highway traffic congestions and travel times, data collection for model calibration.
Traffic data collection to verify macroscopic traffic models and calibrate them to the behavior of local drivers. The data collection uses a computer vision system that track vehicles across a five-lane highway.
URL: /topics/simulation/micro-sensor-management-and-cooperative-sensor-fusion.html
PATH: _topics/simulation/micro-sensor-management-and-cooperative-sensor-fusion.md
Micro-sensor Management and Cooperative Sensor Fusion
Unattended sensor networks may comprise devices of various modalities, such as magnetic, electrostatic, acoustic, seismic and infrared. Sensor nodes may function as a trip-wire, pointer, identifier or tracker. All these nodes contribute to greater situational awareness by fusing the various data sources. At any given time, however, not all nodes are need in the sensor fusion process.
The objective of this project is to develop a self-organized resource management system that allows to nodes to manage their level of activity with respect to data collection, processing and communication in order to reduce power consumption and to extend the lifetime of the network.
URL: /topics/simulation/public-transit.html
PATH: _topics/simulation/public-transit.md
Transportation (Vehicular, Pedestrian, Public Transit)
Plan capacity of public transit systems leveraging computer simulation.
URL: /topics/strategy/data-science-competency-assessment.html
PATH: _topics/strategy/data-science-competency-assessment.md
Data Science Competency Assessment
Data Science Competency Assessment workshops play a vital role in ensuring the successful implementation of AI/ML solutions in data science projects, which in turn leads to optimized day-to-day operations.
One of the key advantages of Data Science Competency Assessment workshops is its ability to bring together key stakeholders responsible for crucial aspects of data science integration. This includes individuals overseeing data sources, accessibility, security, compliance, technical expertise, and compute infrastructure, encompassing both on-premises and cloud environments. By involving these stakeholders, the workshops foster collaboration and enable a comprehensive evaluation of the organization’s data science capabilities.
During the workshop, observations are carefully analyzed to identify the company’s strengths and gaps in terms of data science competency. This analysis is pivotal in understanding the organization’s current state and determining the groundwork required to implement effective data science solutions that address specific business needs. By gaining insights into the strengths and weaknesses, informed decisions can be made to align data science initiatives with the organization’s goals.
Considering the groundwork needed for any data science solution that addresses a business need is of utmost importance. The Data Science Competency Assessment workshop provide a structured approach to assess the organization’s readiness, ensuring that the necessary foundation is in place before embarking on data science projects. By considering the identified strengths and gaps, the workshop facilitates informed decision-making, resource allocation, and strategic planning, ultimately leading to successful implementation and maximum impact.
URL: /topics/strategy/working-backwards.html
PATH: _topics/strategy/working-backwards.md
Working Backwards
Amazon’s Working Backwards method is a unique and innovative approach to product development and decision-making. The method, which has been widely adopted within Amazon, encourages teams to start with the end in mind and work their way backward to determine the steps needed to achieve the desired outcome. This customer-centric approach prioritizes understanding customer needs and preferences before diving into the development process.
At the heart of the Working Backwards method is the creation of a press release and a Frequently Asked Questions (FAQ) document. The process begins by writing a hypothetical press release that captures the essence of the product or feature being envisioned. This press release serves as a concise and compelling description of the product’s benefits, features, and value proposition. It forces teams to clarify their vision and think about how the product will be received by customers and the market.
Following the press release, the team then develops a detailed FAQ document that anticipates questions customers might have about the product. This exercise helps identify potential challenges and provides valuable insights into customer needs, concerns, and expectations. By thoroughly addressing these questions, teams can refine their product idea and make informed decisions about its feasibility and viability.
The Working Backwards method fosters a disciplined and customer-focused approach to product development. It aligns teams around a shared vision and ensures that all stakeholders have a clear understanding of the desired outcome. This approach also promotes collaboration and communication within the team, as members work together to refine the product concept and identify the necessary steps to bring it to life.
URL: /topics/systems/data-provenance-and-workflow-automation.html
PATH: _topics/systems/data-provenance-and-workflow-automation.md
Data Provenance and Workflow Automation
Maintain integrity of knowledge base with provenance tracking of data points generated from multiple processing steps on text documents. Automate workflows for data processing, managing machine-learning models, and scientific computing tasks.