Page 30 - AI Ready – Analysis Towards a Standardized Readiness Framework

AI Ready – Analysis Towards a Standardized Readiness Framework

of procurement-related text documents, mostly in the local language. A natural language
processing model is applied to detect anomalies and flag irregularities in the tender process.

Data collection and model training are based on historical records and data, with expert insight,
from the tender process. After the model is built, anomaly detection is done on the ongoing
procurement process with the help of expert insight, which ensures the verification of the model.
The real-world deployment of a verified model makes sure of the quality of output. The model
will be updated periodically to train on new data such as the latest tender docs so that new
regulations will be included in the training data.

4�5�2 U-Ask

This use case [59] addresses the challenge of finding policy regulations on various government
portals using an AI model trained on United Arab Emirates (UAE) government portal content
and other public sources. Providing a single window of information to the citizens about the
various public service schemes is an important governance initiative. The end users of this
solution are public users accessing the chatbot. The data utilized in this use case comprises
contents from all UAE government portals and other publicly available government sources.
A generative model is employed to produce answers based on personalized requests, while
a prediction and recommendation model is applied to offer precise follow-up questions that
might benefit users. Additionally, a voice-to-text model is implemented to streamline the inquiry
process.

The large language model is trained on UAE government portal content and other government
public sources. The chatbot is trained on queries/responses from the public and citizens. Upon
query from the public, the chatbot generates accurate responses based on the trained model
and context. The operation efficiency and performance of the chatbot is enhanced by the
feedback from users.

4�5�3 Computer Network Fusion Video Brain

This use case [1] combines large models and small models to monitor video content with high
accuracy and flexibility. The large models are used to extract features and infer image events
and behaviours based on colour, texture, shape, and motion; while the small target detection
models then take the task to analyse and predict the content. Cloud-edge collaboration is
required in the process.

The platform offloads the video decoding frame extraction and AI inference service computing
power to the cloud node, realizing the optimal and intelligent scheduling of video analysis
computing resources at the edge side, effectively saving 60% of bandwidth resources and
optimizing the delay by 30%. Expert verification of recognition results is carried out to improve
the accuracy of video intelligent recognition.

Facing the problem of insufficient sample size and data skew in the traditional visual AI training
process, this use case applied artificial intelligence-generated content (AIGC) technology so that
researchers could use large models to produce small samples to improve AI recognition ability.
The deployment of the use case is only available within China Mobile's internal network due to
the use of data from CCTV cameras.

23