Digilytics RevEl’s One-Shot Learning Technology is the Right Way Forward - How One-shot learning is Different from Horizontal OCR?

The electronic prefix ‘e’ has seamlessly attached itself to the United Kingdom (UK) mortgage origination industry. E-mortgage makes processes like application, underwriting, and funding time-saving, accurate, and reliable than the traditional mortgage system.

Mortgage products like?Digilytics RevEl?use artificial intelligence to process and classify documents. It also helps underwriters in making effective decisions based on the borrower’s information.

While?Optical Character Recognition (OCR)?is widely used to scan documents, the?One-Shot Learning (OSL)?technique takes documentation to the next level.

Let’s find out how!

The Necessity for Scanning Documents

Before we get into OCR and OSL processing techniques, it is important to learn the necessity of using these scanning techniques in the first place.

In the traditional loan origination process, the borrower physically submits their application along with all the documents to the lender. The lender or the case handler then manually organises and checks if all the documents are complete with the necessary information.

This step alone can take a long time if the documents or the information is incomplete. Even if all the information is correctly provided by the buyer, manual data processing may lead to many errors.

This is where automated mortgage lending tools come in handy as they can accurately scan the data and check for completeness, correctness, and consistency.

Optical Character Recognition (OCR)

An OCR reader is used to scan and read various numeric and alphanumeric characters from physical documents. This entire process occurs in three steps,

1. The scanner creates a bitmap or image of the document.
2. The software then compares the various characters in the image with its pre-existing database of text and fonts.
3. Once a character is identified, the software converts it into its respective ASCII code which the computer can then understand and display.

As OCR mainly works on pattern recognition, it faces two major drawbacks.

• If a character doesn’t match the OCR software database, it simply skips it and moves on.
• It is difficult for the system to read graphical images and handwritten data efficiently.

While the OCR technique is quite efficient with printed text, these two challenges can cause problems in the mortgage origination process, as every character of data is vital. As a result, case handlers resort to double-checks because of the non-reliability of the extracted information.

The OCR software can be trained to detect all the features and data in a document to make it more efficient and robust. However, training the software may be very time-consuming.

One-Shot Learning (OSL)

Digilytics one-shot learning methods aim at first-time-right applications for mortgage lenders. With OSL algorithms, lenders can accurately process documents and complete more loan contracts effectively in the mortgage lending UK industry.

The Speciality of Digilytics OSL

• Accurate Hybrid Model

The ground-breaking innovation by Digilytics combines the power of?computer vision, machine learning models, and Natural Language Processing (NLP) algorithms.

Being a hybrid model, OSL uses computer vision and deep neural networks to understand and analyse the document with complete precision. This way, it can read and extract information from text, layouts, and images with accuracy levels up to 95%.

• Mortgage specific

Nowadays almost all the NLP models use general English models as bone's model. We are creating our model, with a special vocabulary and syntax understanding for mortgages.

The model is created and trained with mortgage-related terms and regulations.

• Minimal Training

Unlike the OCR method that banks on its dataset and excess training for better results, one-shot learning requires minimal training as it can extract information in “one-shot”. This avoids the need for large datasets and also saves time and resources in continuous training.

One-shot learning algorithms can also check for consistency in the documents by comparing data and signatures with ease.

At Digilytics, we have trained our models and fine-tuned them with over 7 billion mortgage-specific tokens and data from reliable customers.

• Top-Notch Model

Digilytics model incorporates a unique model called MortBERT in its training processes.

While most NLP algorithms focus on the text alone, at Digilytics we train OSL with State-Of-The-Art (SOTA) models that allow document pre-training and understanding in a single framework.

The model uses BERT (Bidirectional Encoder Representations from Transformers) as its base model. BERT is also a SOTA model that is known for its pre-training capabilities.

Advantages and Key Benefits of using the Digilytics OSL Model

1. Pre-trained and Self Learning

The one-shot learning model has already been trained with over 100 documents related to the mortgage origination process. This includes payslips, bank statements, tax returns, and various other forms.

Machine learning mortgage tools and models enable RevEl’s OSL to continuously learn and improve itself while in use.

2. One-stop Destination

RevEl’s one-shot learning capabilities are not solely focused on one particular step of the loan origination system. OSL contributes to every aspect of the United Kingdom’s mortgage origination by delivering accurate results at all times.

Through real-time health checks, OSL technology can also check for the documents' completeness, correctness, and consistency.

3. An Overall Saver

Digilytics OSL model is a saver on all fronts. Its abilities help reduce operational costs and resources by streamlining the mortgage process for both the buyer and the lender. Through digitalisation, RevEl also provides quick mortgage loan clearance and funding

A Proven Model

Based on the live runs with a few lenders in the United Kingdom, RevEl and its one-shot learning AI tech has proved itself to offer absolute value to your money.

These are the results that were immediately seen by the lenders who used this Digilytics product:

• 15% higher gross lending
• 40% increase in funding time
• 30% reduction in operating costs

Conclusion

Digilytics RevEl offers an intuitive and user-friendly?artificial intelligence dashboard?to its users. With One-Shot Learning technology and real-time analytics, document processing and underwriting have never been any simpler.

The OSL model provides better and more accurate results to the lender when compared with its counterpart, the OCR reader. This is due to its mortgage-specific pre-training and self-learning abilities in reading text, layout, and images.

Request a demo with?Digilytics?so that you can also benefit from accurate documentation!

While you are here, other top articles you might be interested in

• All about One-shot Learning AI technology and its impact on the UK Mortgage Market
• A change in technology is coming to the UK Mortgage
• 3C checks: The Digilytics Art of validating mortgage documents
• The Future of Computer Vision, Machine Learning and Artificial Intelligence in Mortgage Industry
• How industry 4.0 principles can work in the favor of mortgage origination?
• Top 5 Real Challenges in Building Predictive Models?in Mortgages
• 5 Ways in which Mortgage Lenders can Leverage Digital Lending for Good
• Whitepaper on Data Lakes: A new approach to managing data
• Applying industry 4.0 principles to mortgages
• Intelligent Digital Income and Expense Verification by Digilytics AI
• Digilytics AI?is one of the top 30 ML-powered start-ups
• Top Challenges in Implementing a Learning Interface
• First time right applications for mortgage lenders – Expectations vs Reality
• Key Takeaways from the Webinar on Intelligent Affordability Services