Tech » Tech update featuring Facebook, Google and IBM

Facebook introduces new model for word embeddings which are resilient to misspellings
Facebook has advanced in the use of natural language processing and is currently implementing a semi-supervised and self-supervised learning techniques to leverage unlabelled data further improving the performance of its machine learning models. Researchers at Facebook recently revealed a new model known as Misspelling Oblivious (word) Embeddings (MOE), which is a combination of fastText and a supervised task which embeds ,misspellings close to their correct variants. MOE’s main aim is to reduce the dealing of malformed words in real-world applications by generating high quality and semantically valid embeddings for misspellings. It surpasses fastText baseline for the word similarity task when misspellings are involved. It maintains the quality of semantic analogies while enhancing the syntactic analogies.

Google AI allows smartphones to interpret sign language
Search engine giant Google’s researchers have rolled a series of algorithms that allow smartphone users to interpret sign language using their phone camera. Google has released the code into the open-source for developers. The sign language uses a combination of a palm detector model known as BlazePalm, a landmark model that works like a fortune teller and a gesture recogniser which can identify hand movements.

IBM trusted AI toolkits for Python combat AI bias
Multinational information technology company IBM’s researchers are finding ways to reduce bias in machine learning models and identify bias in data sets that train AI. The latest toolkit AI Explainability 360 includes algorithms that can be used to explain the decisions made by a machine learning model. The three IBM Trusted AI toolkits for Python are all accessible from GitHub; AI Explainability 360 provides algorithms that cover the explainability of machine learning models and proxy explainability metrics, AI Fairness 360, or AIF360 uncovers metrics to check for unwanted bias in data sets and machine learning models as well as algorithms to mitigate bias and Adversarial Robustness Toolbox a Python library that supports against adversarial attacks making AI systems more trustworthy.

IBM joins Linux Foundation AI to promote open source trusted AI workflows
IBM is aiming to accelerate the development of open-source trusted AI workflows following which the company has become a General member of the Linux Foundation AI (LF AI). The LF AI provides a vendor-neutral space for promotion of AI and ML among others and is backed by major organisations including AT&T, Baidu, Ericsson, Nokia, and Huawei. IBM has plans to introduce AI techniques to all of the foundation’s projects and for that the company will work with the foundation’s committees to create reference architectures and best practices for using open source tools in production.

Experian works in reverse to apply AI, ML to data
Consumer credit reporting agency Experian is working backwards and rethinking their approach in terms of AI and ML adoption. Applying artificial intelligence and automating insight generation is a far cry from traditional data analytics. The infrastructure investments are the foundation for Experian’s data strategy. The effort shows foresight and a commitment from leadership to actually enable an end-to-end innovation cycle” said Santhanam, executive VP and general manager of global analytics and AI. It is eventually going to help to Experian in terms of understanding data outcome requirements whether in improving financial inclusion or building new customer oriented services.

Tap into Amazon.com’s machine learning technology with Amazon Forecast
Amazon has rolled out a new Machine Learning (ML) forecasting tool that uses the same system that runs its shopping platform. The tool uses ML to create custom models that developers can use to make predictions about a host of retail or industrial variables. Forecast heeds product features, seasonality and store locations further using ML to understand any complexities.
The interface has been created with an easy to use API so that developers with different skills can the use the service in AWS. The simplified API lets developers create machine learning models in less than five API calls. Swami Sivasubramanian, Vice President of Amazon Machine Learning commented that: “Amazon Forecast now offers the forecasting expertise from Amazon’s first 25 years of building the world’s largest ecommerce business in a managed service for any company to leverage.

Amazon may be moving into the Cryptocurrency Space
Amazon has developed a managed ledger database Amazon QLDB which can be used to develop blockchain applications. The giant has started their own managed blockchain service for the development of scalable blockchain networks. Amazon has joined forces with other blockchain players including IDEO CoLab, R3, and Consensys. It has also acquired a patent for developing Merkle trees, solutions to proof-of-work algorithm used in blockchain networks like Bitcoin and Ethereum. The company’s capability along with its large customer base may make the company a game changer if it decides to enter the cryptocurrency space.

BOE Governor Mark Carney proposes global cryptocurrency
Bank of England’s Governor Mike Carney has plans to introduce a central bank-issued cryptocurrency like Facebook’s Libra; the Synthetic Hegemonic Currency. He also wishes to improve the structure of the current IMFS (international monetary and financial system) and has asked to consider opportunities presented by advanced technologies. Commenting on the same, Mike Carney said: “Technology has the potential to disrupt the network externalities that prevent the incumbent global reserve currency from being displaced.”

Visa launches security suite
To prevent the threats and frauds around the financial system, Visa has rolled out a suite of new security services and capabilities to detect and disrupt those threats. Among the tools there is Visa Vital Signs that monitors transactions and alerts financial institutions of potential threats at ATMs that may indicate cash out attacks. Another feature uses deep learning to identify hackers who may be taking advantage of the card-less feature by guessing account numbers, expiration dates and security codes through testing. Visa eCommerce Threat Disruption scans the front-end of e-commerce websites for payment day skimming malware.

Blockchain-based securities issuance platform iownit.us greenlighted by Finra
Broker-dealer IOI Capital and Markets (Ioicm) has acquired Finra membership making it easier for the launch of its blockchain-based securities issuance platform. The membership enables Ioicm to act as a replacement agent for digital private securities issues on the blockchain-based-platform -iownit.us. The platform is a brainchild of co-founders Rashad Kurbanov and Hamid Gayibov who were aiming to bring blockchain technology to the private securities market. Built on a private permissioned blockchain that leverages the immutable and secure elements of distributed ledger technology while offering privacy to issuers and investors.