Work-let Name: IOT | Reinforcement Learning based QoS
Worklet
Details 1. Worklet ID: IOTP29PSG
2. College Name: PSG College of Technology
Next Steps
KPIs achieved till now
Key Achievements/ Outcome till now
Any Challenges/ Issues faced
Date: 29.10.2021
● Implemented a pensieve mode that calculates the QOE reward and
selects an optimal bitrate, Trained the model with a data set and
Tested an arbitrary regression model.
● Got an insight on Google Congestion Control Algorithm and
potential use case for implementation in the future
● Created a simple architecture to optimize resolution based on
advanced bitrate calculations. Please refer the diagram and approve
● Issues with our dataset as some of the required parameters were
missing.
● Deploying the ML model, training it and testing it.
● Lack of resources and documentation on required parameters.
● Build a model from scratch that implements our proposed
architecture and try to implement the ml generated equation as the
activation function.
● Investigate the optimization of parameters from our model.
● Work on increasing the efficiency of the model.
● Train and Test the model with a static data set.
● Deploy the model.
● Identified parameters and relevant dataset to be used for the Pensieve
model and its implementation.
● Pensieve model supports bitrate calculations for static videos as they
include the buffer in the dataset.
● The parameters of the Pensieve model and GCC algorithm don’t
overlap since they have different functions by itself and that makes the
integration of the two models impossible.
