Frequently Asked Questions

What is Manot?

Manot is a predictive error analysis platform that allows companies to discover unknown errors and deploy computer vision models with confidence.  Manot helps Product managers and Data Engineers predict and identify potential false positives and false negatives of computer vision models both at pre and post production. Manot helps teams to improve data curation, reduce ML development lifecycle and deliver reliable AI to the real world applications.

How does Manot work?

Manot is a platform that learns from client training data and predicts potential errors for computer vision models before going into production. It analyses and diagnoses CV models' behavior by looking only at the validation set and the model performance on that set. It takes the mentioned as input and builds an embedding set with a novel technique.
At pre-production, either using our proprietary data lake or generative AI module proposes images that could potentially cause that model degradation. The image proposal works on a novel similarity and dissimilarity comparison. 
At post-production, it analyses and proposes images (errors, issues) using customer production data (real-world data; the data that the model sees while operating in the production enviroment).

What is bias/outlier/edge case fimin the computer vision model?

Bias in the computer vision model refers to the presence of systematic errors or inaccuracies in the model's predictions, which can result in unfair or discriminatory outcomes. It is an image with patterns and objects that have not considered to happen with the model when it is deployed in the real world; i.e. it is outside of the training data distribution.

What are the reasons leading to bias?

Bias can arise from various factors, such as the data used to train the model, the algorithms and techniques used to develop the model, and the context in which the model is used.

How does Manot detect hidden biases and outliers in models?

With its advanced scoring system, Manot takes your training data and real-world data and identifies the areas where your model underperforms, detects hidden biases, and suggests ready data samples that will contribute the most to rebalancing the training dataset.

What is the meaning of computer vision model ethics? 

Computer vision model ethics refers to the ethical considerations and principles that should be taken into account when developing and deploying computer vision models.