Anthill Inside 2019

A conference on AI and Deep Learning

Tickets Propose a session

Network health predictions and optimization recommendation using Deep learning Neural network models and Reinforcement learning

Submitted by Anuradha K (@anuradhak) on Tuesday, 7 May 2019


Preview video

Section: Workshops Technical level: Intermediate Session type: Workshop

Abstract

Time series prediction of network parameters and detecting network health with network performance optimization, has been an interesting problem to solve for researchers in the field of Machine Learning and Data Mining community. These use cases are present across different industries like retail, telecom, transport with good presence in Telecom industry. However, there remains a challenge in getting a good prediction accuracy and efficiency while solving these problems. Traditional approaches typically involve extracting discriminative features from the original time series using dynamic time warping (DTW) or shapelet transformation, and traditional ML models are applied on top of these transformations to get decent accuracy. These methods are mostly ad-hoc and the performance of these models are limited as there is a separate process to extract features and another process to predict. Recommending an optimal parameters for network is normally done by training more data in traditional supervised models. There lies a challenge in the supervised learning models as these models are data hungry. If the data is insufficient, the traditional supervised models fail to converge, and mining patterns in the data can be a challenge. To address the first challenge, we propose end-to-end neural network architecture models such as Univariate/Multivariate-LSTM,  LSTM - Convolutional Neural Network , CNN, LSTM-CNN which incorporates feature extraction and prediction in a single framework. To address the second challenge, Deep Reinforcement learning has been used to recommend the optimal parameters with predicted network parameters which in turn can lead to good network health. We did comprehensive empirical evaluation with various proposed methods on a large number of benchmark datasets, the approach based on Deep learning neural network models and Deep reinforcement learning methods in network parameter optimization has provided a good accuracy when compared to the existing models.

Outline

INTRODUCTION
AUTHORS INFORMATION
PROBLEM STATEMENT/CONTEXT
EXISTING TOOLS/SOLUTIONS
PROPOSED SOLUTION
BENEFITS OF OUR PROPOSED SOLUTION
CURRENT SOLUTION SCENARIO
PROPOSED SOLUTION SCENARIO
DIFFERENT MODELS /APPROACHES AND RESULTS
CNN LSTM ARCHITECTURE
COMPLETE PROPOSED SOLUTION ARCHITECTURE
TRADE-OFFS IN PROCESS FOR SOLUTION
PRIVACY, REGULATORY AND ETHICAL CONSIDERATIONS FOR THE DESIGNED SOLUTION
KEY TAKEAWAYS
ADDITIONAL DETAILS THAT WILL BE SHARED DURING PRESENTATION
QUESTIONS???

Requirements

Anaconda,python installation to be completed in laptop

Speaker bio

https://www.linkedin.com/in/anuradha-karuppasamy-b5378813b/

I am a Senior Data Scientist from Ericsson GAIA (R&D). I have 18+ Years of experience in the areas of Telecom, Retail services, Sales and transportations, Financial and Banking and delivered projects of various sizes. I have played several roles like Senior data scientist, AI technology Architect Delivery Manager, Analytics Tech lead, Data scientist, Data Statistical Analyst, Solution Project Manager, Technical/Project lead and developer. My key Capabilities are Machine learning, Analytics, Artificial intelligence, Deep learning, Reinforcement learning, Python, IBM Watson, Azure ML, Biometrics (Facial and Emotion recognition), Predictive Modeling, and Speech to Text.

Slides

https://drive.google.com/file/d/1ZtFsyuW6Cm5m112YZPYDTco8WbBUpSVK/view?usp=sharing

Preview video

https://photos.google.com/share/AF1QipP1lnLzatLkLGcHO1gnE1aD6fXL1uN2pIV8wr83_4m2iF-JhMD5i61qJmr3Ivc_kg/photo/AF1QipPIYIljnrWP9QtvKWdey3pWevShoZrErfJDiT5t?key=UU1KMmRGLW9OVHdQVFVCT1JmWGxZcGZXcGpIZVR3

Comments

  • kumar vishwesh (@khiladikumarvishwesh) 5 months ago

    I can spot a lot of utility in the paper. In real-time use cases, this innovative approach has potential to automate intelligently with strong business impact. A good read on AI/ ML.

  • Arvind Dangi 5 months ago

    Brilliant paper. A nice read.

  • Sachin Upadhyay (@100492300638) 5 months ago

    good to see it

  • Praveen kumar 5 months ago

    A practical paper with full of innovation.

  • pawan kumar Kurmi 5 months ago

    Awesome and unique, veey different from classical approaches.

  • Ayush Pradhan 5 months ago

    Interesting paper.

  • smitha kannur 5 months ago

    Nice read.

  • Priyajit Nanda (@priyajitnanda) 5 months ago (edited 5 months ago)

    Genius .....Such a innovative display of ideas

  • Abhishek Balaji (@booleanbalaji) Reviewer 5 months ago

    Hi Anuradha,

    Thank you for your proposal. It is not clear if this proposal is a talk or a workshop/tutorial, given the requriements. Please edit your proposal to update the same. The domain the proposal falls under - Deep Reinforcement learning ties in with Anthil Inside 2019 (https://hasgeek.com/anthillinside/2019) more than The Fifth Elephant. Further, the slides are very thin right now and we’d need more details to evaluate the proposal. Your slides must contain the following:

    1. Problem statement/context which the audience can understand and relate to. This has to be a problem present in the industry and cannot be specific to your company
    2. What are the tools/options available in the market for this problem? How did you evaluate these, and what metrics did you use for the evaluation? Why did you decide to build your own model?
    3. Why did you pick the option that you did?
    4. Explain how the situation was before the solution you picked/built and how was the fraud/ghosting after implementing the solution you picked and built? Show before-after scenario comparisons & metrics.
    5. What compromises/trade-offs did you have to make in this process?
    6. What are the privacy, regulatory and ethical considerations when building this solution?
    7. What is the one takeaway that you want participants to go back with at the end of this talk? What is it that participants should learn/be cautious about when solving similar problems?

    We’d need to see the updated slides on or before 21 May, for us to evaluate your proposal. If these details are not updated, we’d only be able to consider the proposal for a future event.

    • Anuradha K (@anuradhak) Proposer 5 months ago

      Hi Abhishek, I have sent my updated slides to your id. Please let me know if you have received it

    • Anuradha K (@anuradhak) Proposer 5 months ago

      Sure Abhishek Balaji. Thanks for your time. I will definitely update the slides as soon as possible (Before 21st May).

  • Aditya Patel (@adityap) Reviewer 4 months ago

    Hi Anuradha,
    Great application of deep learning.But i am a bit confused, Since you have mentioned Reinforcement Learning but have not explained it anywhere. Most of the slides are around hyper-parameter tuning of Deep Neural Networks. Also couldnt find any results to prove the efficacy of the model. Can you please elaborate on these points?

    • Anuradha K (@anuradhak) Proposer 4 months ago

      Thanks Aditya for your review and time. I have updated the information as requested in slides
      (slide numbers: 12,13,14) and also updated the slide link. Kindly let me know if anything else is required. Thanks.

      • Aditya Patel (@adityap) Reviewer 4 months ago

        Hi Anuradha,
        Thank you for the edits. Can you also add a background on Reinforcement Learning. Since it’s a workshop, you might want to start off with a generic example.

        • Anuradha K (@anuradhak) Proposer 4 months ago

          Thanks Aditya for your time and review. I have udpated the required information as requested in slides (slide number: 14). Please let me know. Thanks.

Login with Twitter or Google to leave a comment