Blog

It is said in literature that« sometimes we survive by forgetting ».During the development and progress of AI and Machine intelligence, this errand becomes more vibrant. We find typical machine learning models which fail in continuous learning. This is an inclination towards forgetting certain new trends, when exposed for training or learning new tasks especially to the neural networks or even to the human brain. This is usually referred to as catastrophic forgetting. It is realistic to justify such behavior of an intelligent system, where the distribution of inputs across different new tasks may occur very randomly.

In recent years, global attention has not been drawn to this all-important AI phenomena i.e. forgetting. For any stream of the continual learning setting, we consider that the learning process should be stochastically optimized with a stipulated timeline. Normally, it is known in Stochastic Gradient Descent (SGD) optimization, where any snap of tasks can be referred to as a tiny “task” presented to the network as new a task to learn. Lets take an example from the perspective of MUST matchmaking platform: For optimal matching purpose, MUST follows a supervised model to train the system. The supervised model comprises of different data features of several companies like their production, categories, supply details to name a few. If we consider a fixed number of attributes to train our model, it becomes realistic. However, the real challenge is that these features are dynamically changing and seldom there is practical need to introduce new features and new tasks or processes to learn for an optimal matching of networkers. In terms of time optimization, good learning means that the model can well recall the classification at a time t, 0 < t. It fails if the available data is not optimized to handle these new tasks and features afterwards.

At this point, for MUST matchmaking, we introduce multitasking procedures while presenting data features with well-defined stochastic gradient optimization. This will be adequate to address this miss classification of tasks and features in a more generic way.

As a result, end users will enjoy the exact matching partners without crossing over despite of the dynamic features of the data, never to be forgotten by the AI model.

The focal points of this investigation depends on:

· To keep an appropriate heuristic weight to each new functionality added for enterprises for training.

· To prioritize the precedence of tasks and features for optimal matchmaking

· Finally, a smart optimization for the volume of data steams to be processed by Must platform towards a scalable solution .

Thus with Must Metaverse platform , the participants can be extended to perform following tasks while AI/ ML as a lead component:

• Analysing customer data to understand their needs and desires
• Developing prototypes of new products and services
• Creating realistic simulations of new products and services if required through Must Pavilion
• Testing new products and services before they are released.
• Share and interact best Matchmaking practices

Must Metaverse on the way towards creating a "Digital Human Platform" of future

At Must we always believe that forgetting is not acceptable to any of the end-users, so we follow the age-old saying by Nelson Mandela " I never lose or forget , I either win or learn”…