top of page
MAS methodologies

Figure 1: Evaluation of agent methodologies for projects.  Source: Himanshu Patel, Riga Technical University, Evaluation of methodologies for efficient multi-agent System projects, 2017

Services for Airport using GAIA

Figure 2: MAS system for airport ( adopted from Sanchez-Pi, Carbo et. al, 2010)

Source: Himanshu Patel, Riga Technical University, Evaluation of methodologies for efficient multi-agent System projects, 2017

Project Title / Status: Evolution Of Methodologies For Efficient Multi-Agent System Projects / Completed 

​

Our contribution: We delivered our expertise in technical thesis writing, external mentoring and ensured our client had best learning experience, ease of learning and assisted our client in becoming more productive. We offered this as part of our services package

​

Project methodology:

Based on proposed requirements, we supported our client’s need with a self-defense presentation and offered supervision of thesis writing. We delivered a high quality material based on our vast experience resulting in approval by university guides with an A grade. We tested the originality of the content using anti-plagiarism tools that ensured our content did not conflict with other authors

​

Evaluation of agent methodologies are shown in Figure 1. We designed tool for MAS methodology selection using a mini-expert system, AI / fuzzy logic, and statistics. Java based tool was hosted in AWS and used Spark.

​

Case study:
We developed case study details related to design of services for an airport using GAIA. We identified different agent types which are: Client, Provider and Central agents. The Central agent depicted the infrastructure which captured location information from various sensors. The Provider agent acted on behalf of various services; Client agent denoted users with wireless devices. Models like Environmental Model, Role Model, Interaction Model & Service models were developed that ensured the demonstration of the concept.These are shown in Figure 2

Location aware intelligent system

Figure 3: Architecture of SpatialAgent  framework ( adopted from Ichiro Satoh, National Institute of Informatics, Japan, 2005)

RFID for location aware system

Figure 4: RFID technology based wrist bands (left) and USB desktop reader (right), ( adopted from Dante. I. Tapia et.al, Universidad de Salamancha, Spain, 2008)

Location aware system App

Figure 5: Positions of RF Tags in floor and map-viewer agent in a workstation ( adopted from Ichiro Satoh, National Institute of Informatics, Japan, 2005)

Project Title / Status: Location aware ambient intelligent system / Completed

​

​Our contribution: We delivered our expertise in technical project planning for ​location-aware system in which spatial regions were determined within a few square feet, so that one or more portions of a room or building was distinguished using RFID tags  enabling a physical entity and place to spatially bind with one or more mobile agent-based services. These services annotate and support the entities or places that made the services to be dynamically deployed at stationary and mobile computing devices

​

Project details:

This framework provides the middle-ware infrastructure for managing location-sensing systems and deploying mobile agents at suitable computing devices according to the locations of users and objects, including computing devices. It consists of three parts: (1) location information servers, called LISs, (2) service-provider mobile agents, and (3) agent hosts, as we can see in Figure 3.

 

We used RFID transponders, sensors and central PC along with ZigBee standards based boards, configured in a mesh network with various roles like controller, sender/receiver/ repeater as in Figure 4. Ambient and location based data captured by sensors are uploaded and agents installed on the systems take appropriate actions like switching off lights, beeps, alerts and various programmed actions

​

Use Cases:

The first example allows us to use a PDA to remotely control nearby electric lights in a room

​

The second example offers a user assistant agent that follows its user and maintains profile information about him/her inside itself, so that he/she can always assist the agent in a personalized manner anywhere

​

Third example is user navigation system ( Figure 5) that assists visitors to a building. The location information servers (LIS) detects the place-bound agent that is tied to the tag. It then instructs the agent to migrate to its agent host and provide the agent’s location-dependent services at the host

ML Pipeline
Random Forest

Project Title / Status: Predicting Grant Applications / Completed

​

​Our contribution: We delivered our expertise in technical project implementation for  â€‹predicting the grant applications using spark framework.

​

Project details:

We demonstrated use of various functions available in Spark’s machine learning (ML) libraries in solving real world problems by creating features, building a pipeline, cross validation and model tuning that helped to predict the outcome of grant application by various scholars.

  • Basic setup of spark notebook

  • Used  Spark csv package from databricks to load csv data file

  • Created features for given datasets by re-encoding some data using withColumn on dataframe

  • Converted features into vectors for use in spark’s estimators 

  • Setup a RandomForest classifier using spark’s ML library

  • Built a pipeline to chain multiple spark transformers to create a workflow

  • Used the transformer to index and assemble for the output data frame

  • Created a binary classification evaluator

  • Split the training and test dataset

  • Fitted the model and made predictions

  • Tested to avoid overfitting by using cross- validation when training the model

  • Improved the model’s performance by using grid search to find better parameters

  • Extracted useful information from the results of the grid search, including:

    • the average area under the ROC curve for each combination of parameters

    • the feature importance of the best model

Big Data Computing model
Airline delay prediction

Project Title / Status: Big Data Analysis for Airline Delays / Completed

​​Our contribution: We delivered our expertise in technical project implementation for  ​predicting the airline delays for arrivals and departures  using Revolution R, and a publicly available data set.

Project details:

Developed model using rxLinModel, use of R functions like rxSummary, rxQuantile, rxCrossTabs, rxCube to analyze aircraft delays and the relationships w.r.t the weekdays, holidays and any deviations from standard schedules. Used transformations for dataframe to XDF format and use of Transforms/ TransformFunc for data manipulation and display to verify the changes

  • Set up the local and cluster environment – Microsoft HPC, Microsoft Azure Burst, LSF

  • Defined Compute context using  RevoScaleR ComputeContext for defining hardware, monitoring, and failover

  • Imported Dataset and scaled down to a small subset 

  • Summarized variables like 7 days of week category

  • Used Linear model to predict arrival delay

  • Summarized the model to see its performance 

  • Used advanced features like rxLinMod, rxPredict

  • Predicted using advanced features

  • Re-trained model , cross-validated model and retested the model

  • Observed the model performance and shown the delay predictions were accurate

bottom of page