Differential Privacy

• Key player: Microsoft leads the differential privacy field with a score of 144.6, according to emerging technology blog Linknovate. Its nearest competitor is Apple with a score of 101.5. It is significantly involved in the field's development.
• Products and solutions in differential privacy: Microsoft is actively working with its partners (including Harvard) to develop open toolkits "to better enable differential privacy." The first open source platform of its kind was announced in June 2020; it uses Harvard's Open DP initiative.
• Products and solutions in other fields:
• Encrypted deep learning: Microsoft offers Microsoft Simple Encrypted Arithmetic Library (SEAL) to facilitate encrypted computations. This service was established on March 27, 2016.
• Model integrity: Microsoft offers Azure for natural language processing solutions. Azure became available on February 1st, 2010.
• Synthetic data: The primary solution Microsoft offers for data simulation is Azure, which launched on February 1st, 2010.
• Federated learning: Microsoft released research in March 2020 related to its in-developed federated learning framework.
• Format preserving encryption solutions: Microsoft does not offer a first-party format-preserving encryption solution.
• Related Findings: As of 2018, the most active organizations in the field of differential privacy are SMEs and startups with 43 percent of activity. Universities are not far behind with 36 percent of activity in the field.
• Related finding: Of the top 10 organizations involved in differential privacy, the top three are tech corporations based in America. Of the remaining seven, six are American universities (and one is a Chinese university).

Encrypted Deep Learning

• Key player: Intel in engaged in creating Homomorphic Encryption technologies that are more accessible and can scale up. The possibility of widespread application is significant and makes them a key player.
• Products and solutions in encrypted deep learning: Its HE Transformer for nGraph is a backend to Intel's nGraph, a "graph compiler for Artificial Neural Networks." This solution became available in November 2019.
• Products and solutions in other fields:
• Differential privacy : Intel does not have a differential privacy solution but produced a related white paper on AI privacy in 2018. 
• Model integrity : Intel offers NLP Architect for natural language processing. This became available in 2018.
• Synthetic data : Intel does not provide a synthetic data solution.
• Federated learning : Intel has partnered with Data Republic to develop and offer federated data solutions to financial institutions. The first trial of Intel's solution was completed in 2018 with Singapore's United Overseas Bank.
• Format preserving encryption solutions: Intel does not have a software-based format-preserving encryption solution.

Model Integrity

• Key player: IBM is a key player in the model integrity or natural language processing (NLP) field. This is mainly due to the prevalence and dominance of its Watson AI.

• Products and solutions in model integrity: IBM's Watson is the primary offering in NLP solutions.
• Products and solutions in other fields:
• Differential privacy : IBM released its open source differential privacy tool, the IBM Differential Privacy Library, in June 2020.
• Encrypted deep learning : IBM completed its first successful field trials of "fully homomorphic encryption" to enable encrypted deep learning in July 2020.
• Synthetic data : IBM InfoSphere Optim Test Data Fabrication enables the production of synthetic data.
• Federated learning : IBM's Federated Learning tool was released in July 2020.
• Format preserving encryption solutions: IBM hosts an FPE Translate callable service for some format preserving encryption purposes. It is not clear how long it has been available.

Synthetic Data

• Key player: Mostly AI is a key player in the synthetic data segment. This is due to the significant interest and investment it has received from major banks and partners since its launch, including one of the top five insurance companies in the U.S.

• Products and solutions in synthetic data: Mostly Generate is Mostly AI's offering in the synthetic data field; it has been available since 2017.
• Products and solutions in other fields:
• Differential privacy : Mostly AI does not have a differential privacy solution but one of its staff wrote a recent blog post on it.
• Encrypted deep learning : Mostly AI does not offer encrypted deep learning but its CEO was recently interviewed and discussed it.
• Model integrity : Mostly AI does not offer model integrity solutions.
• Federated learning : Mostly AI does not offer federated learning solutions, but one of its staff was interviewed recently and discussed its relationship with synthetic data.
• Format-preserving encryption solutions: Mostly AI does not offer a format-preserving encryption solution.

Research Strategy

Machine Learning in the Finance Sector

• Machine learning (ML) privacy-preserving solutions have the power to deliver businesses value and impact across different industries, such as the finance industry, which has particularly embraced ML privacy-preserving solutions by investing predominantly to monetize data assets through these solutions.
• Banks such as J.P. Morgan, Wells Fargo, Bank of America, City Bank are already using machine learning, preserving solutions to provide varying facilities to customers looking for proactive risk prevention and detection. Just like the in the U.S., banking institutions in India are embracing privacy-preserving machine learning solutions to improve not only customer experience and customer data management, but also fraud detection in real life, risk modeling for investments.
• A recent survey conducted by the Bank of England and Financial Conduct Authority shows that more than two-thirds of financial service organizations in the U.K. have implemented a privacy-preserving machine learning application. Further, the study shows that the usage of privacy-preserving machine learning in financial institutions in the U.K. is likely to double in the next three years.
•  Regulatory and cybersecurity issues concerning data have significantly affected the finance industry. While the finance industry is in dire need to incorporate machine learning as a proactive solution, A.I. researchers are constrained by data quality and availability. For instance, there has been a need to identify databases that would enable them to be transparent on things such as stamp out financial fraud, which is estimated to be a $5 trillion global issue. Innocuous datasets such as ImageNet have seen machine learning speed up advances as they are freely available.
• In 2018, through a $10 million financing, J.P. Morgan partnered with data security and privacy-preserving machine learning provider, Inpher. Inpher Inc., a renowned global data security and analytics company based in New York and Switzerland, has pioneered an advanced cryptographic Secret Computing platform. By partnering with J.P. Morgan, the latter stood to benefit from data protection while being processed, privacy-compliant analytics that observe new laws, and distribution of data sources without exposing any information.
• Institutions such as Wells Fargo have not publicized their A.I. initiatives. Through its machine learning service provider, Feedzai, Citibank has made significant strides in A.I., particularly in countering fraud and money laundering through privacy-preserving machine learning solutions. Citi Bank partnered with Feedzai to strategically counter mishaps caused by privacy breach by integrating a fraud detection software to counter such activities by monitoring customer payment behavior and sending n alarm in case of deviations such as unusually large funds transfers.
• The Security and Exchange Commission’s Privacy of Consumer Financial Information Rule expects financial institutions to protect their clients’ data even as data-hungry tech giants continue to look for ways to gain this private data.
• According to studies, the banking industry lost $2.2 billion in fraud in 2016. These losses catapulted the need for privacy-preserving machine solutions, mainly because 58% of these fraud cases were debit related. It was predicted that card fraud cases would increase by 42% by 2020, but as of 2018, payment fraud cases in the U.S. decreased to $1.8 billion. Machine learning privacy-preserving solutions have been beneficial to the financial sector by detecting fraud or breach of data and helping financial institutions take proactive measures.