Feedzai Data Scientist Wins National Pattern Recognition Award
Latest posts by amit sonawane (see all)
- Friendly Fire: Dealing with Deliberate Chargebacks from Consumers - May 19, 2016
- The future of payments and what it means for fraud - May 3, 2016
- The Internet of Things presents new vectors for payments – and fraud - March 24, 2016
How many of us have given a 5-star rating online to review a seller, product or service on-line? Or classified an e-mail message as spam? What many of us did not know is, in that moment, our rating fine-tuned a Machine Learning model. Human feedback bridges Artificial Intelligence last mile to our world. Each pooled opinion about Artificial Intelligence decisions improves the general customer experience – even across vendors. Every AI decision thereafter benefits from that human feedback.
This technology is heralded as the most effective tool keeping financial transactions frontlines safe.
Companies who wish to stay on the leading edge of AI powered Fraud Prevention are always on the lookout for the best and brightest additions to their teams. Feedzai boasts a world class team of industrious software engineers and focused data scientists who develop a leading Omnichannel Fraud Prevention solution.
Mariana Lourenço is a young data scientist with an excellent academic track record, working in Feedzai’s Lisbon office. Mariana graduated with a Masters in Informatics Engineering, (specialized in Intelligent Systems) in the top 3% of her class at University of Coimbra, Portugal.
Feedzai is proud to announce Mariana’s Masters Thesis, titled “Supervised Topic Models with Multiple Annotators” has been awarded the Best Master Thesis 2015 in the field of pattern recognition, by the Portuguese Association for Pattern Recognition.
In the course of her research, Mariana developed novel ways to improve the overall reliability of an Artificial Intelligence algorithm that crowdsources indiscriminate opinions (feedback) about its Machine Learning models decisions quality. Her enhanced algorithm derives the main subject matter of text articles.
Common Artificial Intelligence algorithms place excessive trust in human supervising feedback. Machine algorithms frequently treat every feedback input as valuable, even while not everybody inputs accurate decisions. People can mistakenly flag legitimate e-mails as spam, and may write unjustified reviews at on-line marketplaces. Erratic feedback can negatively impact Artificial Intelligence algorithms internal calibration and runtime accuracy.
Mariana’s contributions to the Artificial Intelligence field improves the state-of-the-art sLDA algorithm by identifying emerging human experts and accurate appraisers among the crowd fine-tuning Machine Learning model decisions. Mariana demonstrated that people who provide reliable feedback deserve increased influence above careless people when evaluating Machine Learning model decisions.
Mariana’s innovative MA-sLDA algorithm detects the subject matter of text articles automatically up to 90% of times. This result represents a double-digit efficiency improvement, 10% on average, over state-of-the-art sLDA algorithm. Her MA-sLDA approach was particularly strong processing unreliable information.
Whereas sLDA works in similar fashion to an Encyclopedia Britannica expert, Mariana’s “Multiple Annotator – Supervised latent Dirichlet allocation for classification” algorithm is closer to the Wikipedia model. Wisdom of the crowd is availed to achieve the best results.
Her trailblazing Machine Learning research earned Mariana an invited research scientist position at Singapore – MIT Alliance for Research & Technology (SMART). There, she employed a MedLDA algorithm to estimate the date and location of mass gatherings of people, using just snippets of text, crowdsourced from low fidelity Google results. Her research improved the frequency of public buses and satisfied peaks of demand around large events. The algorithm predicted attendance volumes and managed to avoid empty buses on less popular events. It paved the way for smart cities to conserve energy and effectively manage limited public resources with the support of Artificial Intelligence.
Feedzai’s team grows stronger with bright professionals like Mariana, capable of developing next-generation Real-Time Omnichannel Fraud Prevention solutions.
Feedzai Fraud Prevention solution features Segment of One profiles to provide historical card and customer behavior to transactional data. Data Enrichment gathers crowdsourced relevant facts from social media and beyond. The wealth of syncopated data enables Feedzai to perform informed best-in-class transaction decisions. Whitebox Scoring presents fraud analysts with clear text explanations for each Risk Score.
Fraud Analysts go over each unclear transaction lined up for Review. The investigation surrounding the true nature of each payment under review culminates in a decision allowing or blocking it. That choice has a positive impact on future Machine Learning model decisions.
Mariana’s work has the potential to reward the most accurate Fraud Analysts payment verification judgements, increasing their overall influence on Feedzai’s Fraud Prevention Machine Learning models. Mariana’s groundbreaking work promises to takes Artificial Intelligence one step closer to perfect decisioning.
Feedzai is spearheading Fraud Prevention using Real Time Machine Learning at Big Data scale to stop today’s and tomorrow’s fraud rings. Mariana Lourenço research moves the needle on Artificial Intelligence accuracy ensuring Feedzai on-premise and cloud products don’t rest on their laurels. Feedzai aims to keep driving innovation forward.