Periodic Reporting for period 1 - IA for OIP (Innovation Associate for OIP)
Reporting period: 2021-01-01 to 2021-12-31
For many people, TV and video are vital for their daily lives. Not only as source of entertainment, but also for their daily news. The Covid pandemic has shown how critical this way communicating is.
Today, there are many ways for consumers to watch video and TV and just as many ways for the video content to reach them. One method is broadcasting, where content is sent out to everyone simultaneously. Another method is direct transmission over the internet. These methods require numerous systems and technology components to work together to deliver the content. These include different functionalities like recording, watching it later and or watching it on demand. Unfortunately, each of these systems can fail and impact consumers’ viewing experience negatively. Currently, resolving these impacting issues can take several days and, in the meantime, viewers suffer from a reduced viewing experience. We want to ensure that those consumer-facing issues have as minimal impact as possible.
This is why our mission is to ensure video content quality (with as little failures as possible). We provide operational services for critical delivery infrastructure and solve video delivery problems quickly. To proactively detect and solve failures fast and improve response/resolve times, we explore automated ways of detecting and analyzing failures and underlying video delivery problems. We use machine learning to automatically detect anomalies in a complex video delivery environment and identify failure with their root causes.
Today, there are many ways for consumers to watch video and TV and just as many ways for the video content to reach them. One method is broadcasting, where content is sent out to everyone simultaneously. Another method is direct transmission over the internet. These methods require numerous systems and technology components to work together to deliver the content. These include different functionalities like recording, watching it later and or watching it on demand. Unfortunately, each of these systems can fail and impact consumers’ viewing experience negatively. Currently, resolving these impacting issues can take several days and, in the meantime, viewers suffer from a reduced viewing experience. We want to ensure that those consumer-facing issues have as minimal impact as possible.
This is why our mission is to ensure video content quality (with as little failures as possible). We provide operational services for critical delivery infrastructure and solve video delivery problems quickly. To proactively detect and solve failures fast and improve response/resolve times, we explore automated ways of detecting and analyzing failures and underlying video delivery problems. We use machine learning to automatically detect anomalies in a complex video delivery environment and identify failure with their root causes.
To adequately address the market needs, Divitel will offer a plug-and-play solution to provide video delivery operators direct insights into their operations without a lengthy onboarding and configuration phase. We achieve this by sharing knowledge between deployments without exposing anything about the consumer. We use this knowledge to train machine learning models to detect potential incidents. There were three phases to our work:
1. Drawing from our extensive video domain knowledge, we explored different video delivery architectures. Here we identified components, connections between components, transport of information, and data sources for us to use in machine learning algorithms. We included entire end-to-end paths and the dependencies between parts to make precise predictions.
2. We designed a data architecture to support the data extraction, processing, storage, and loading of data for further use. We designed a pipeline that starts with unstructured log data and finishes with highly structured tabular data. We then use and analyze this data and provide initial insights, potential solutions through our engineers, or customer explanations. This data is also the basis for our next phase.
3. We developed machine learning models for fast and accurate incident detection that support us in providing excellent operational services. We use and deploy these models and instantly create insights on a customer deployment through our proprietary way of sharing knowledge. The models detect anything from slight variations that signal a potential impact that a human operator would overlook, to significant and directly impacting events.
Divitel will build this data platform based on the derived architecture following the steps mentioned above, thereby creating highly automated operational services. We will use and improve the machine learning models in this proof-of-concept and refine knowledge sharing between deployments for even better results. We are also in the process of applying for several patents derived from this project.
1. Drawing from our extensive video domain knowledge, we explored different video delivery architectures. Here we identified components, connections between components, transport of information, and data sources for us to use in machine learning algorithms. We included entire end-to-end paths and the dependencies between parts to make precise predictions.
2. We designed a data architecture to support the data extraction, processing, storage, and loading of data for further use. We designed a pipeline that starts with unstructured log data and finishes with highly structured tabular data. We then use and analyze this data and provide initial insights, potential solutions through our engineers, or customer explanations. This data is also the basis for our next phase.
3. We developed machine learning models for fast and accurate incident detection that support us in providing excellent operational services. We use and deploy these models and instantly create insights on a customer deployment through our proprietary way of sharing knowledge. The models detect anything from slight variations that signal a potential impact that a human operator would overlook, to significant and directly impacting events.
Divitel will build this data platform based on the derived architecture following the steps mentioned above, thereby creating highly automated operational services. We will use and improve the machine learning models in this proof-of-concept and refine knowledge sharing between deployments for even better results. We are also in the process of applying for several patents derived from this project.
In a complex landscape such as video delivery, failures that greatly affect viewers cannot entirely be eradicated. Video delivery service operators still operate primarily manually, negatively impacting consumer viewing satisfaction (not fast enough to fix failures). This is why we strive for immediate or even predictive detection to minimize the time and the extent a customer is affected.
This project was our first step towards this goal. We can detect incidents in near real-time and immediately act on them without impacting the consumer, achieving high customer satisfaction. We created the foundation for our data platform through the design and proof-of-concept implementation of the data pipeline. Knowledge sharing across deployments also enables us to service many different video service operators without complicated onboarding processes or extensive deployment analysis, leading to faster deployments and lower up-front costs.
Lower costs lead to a chance for smaller operators to enter or remain relevant in the market and provide a wider diversity of offers, including for subscriber bases that would not be large enough to be profitable to market to without it. In the future, our approach can also be generalized to move beyond the video delivery industry to any other domain which exhibits similar complexities of distributed components over a network and a fragmented technology ecosystem.
This project was our first step towards this goal. We can detect incidents in near real-time and immediately act on them without impacting the consumer, achieving high customer satisfaction. We created the foundation for our data platform through the design and proof-of-concept implementation of the data pipeline. Knowledge sharing across deployments also enables us to service many different video service operators without complicated onboarding processes or extensive deployment analysis, leading to faster deployments and lower up-front costs.
Lower costs lead to a chance for smaller operators to enter or remain relevant in the market and provide a wider diversity of offers, including for subscriber bases that would not be large enough to be profitable to market to without it. In the future, our approach can also be generalized to move beyond the video delivery industry to any other domain which exhibits similar complexities of distributed components over a network and a fragmented technology ecosystem.