Yassine LASRI (@ylasri) header
Yassine LASRI (@ylasri) profile picture

Product Manager at @SynaptiqueM, IT & Telecoms Professional, Interested in AI, IoT, 5G, Big Data, RA, FMS, NG B/OSS, Mobile Money

Morocco
Joined January 2009
Photo by Yassine LASRI (@ylasri)Photo by Yassine LASRI (@ylasri)Photo by Yassine LASRI (@ylasri)Photo by Yassine LASRI (@ylasri)Photo by Yassine LASRI (@ylasri)Photo by Yassine LASRI (@ylasri)Photo by Yassine LASRI (@ylasri)Photo by Yassine LASRI (@ylasri)Photo by Yassine LASRI (@ylasri)Photo by Yassine LASRI (@ylasri)
Yassine LASRI retweeted
TL;DR: We built a transformer-based payments foundation model. It works. For years, Stripe has been using machine learning models trained on discrete features (BIN, zip, payment method, etc.) to improve our products for users. And these feature-by-feature efforts have worked well: +15% conversion, -30% fraud. But these models have limitations. We have to select (and therefore constrain) the features considered by the model. And each model requires task-specific training: for authorization, for fraud, for disputes, and so on. Given the learning power of generalized transformer architectures, we wondered whether an LLM-style approach could work here. It wasn’t obvious that it would—payments is like language in some ways (structural patterns similar to syntax and semantics, temporally sequential) and extremely unlike language in others (fewer distinct ‘tokens’, contextual sparsity, fewer organizing principles akin to grammatical rules). So we built a payments foundation model—a self-supervised network that learns dense, general-purpose vectors for every transaction, much like a language model embeds words. Trained on tens of billions of transactions, it distills each charge’s key signals into a single, versatile embedding. You can think of the result as a vast distribution of payments in a high-dimensional vector space. The location of each embedding captures rich data, including how different elements relate to each other. Payments that share similarities naturally cluster together: transactions from the same card issuer are positioned closer together, those from the same bank even closer, and those sharing the same email address are nearly identical. These rich embeddings make it significantly easier to spot nuanced, adversarial patterns of transactions; and to build more accurate classifiers based on both the features of an individual payment and its relationship to other payments in the sequence. Take card-testing. Over the past couple of years traditional ML approaches (engineering new features, labeling emerging attack patterns, rapidly retraining our models) have reduced card testing for users on Stripe by 80%. But the most sophisticated card testers hide novel attack patterns in the volumes of the largest companies, so they’re hard to spot with these methods. We built a classifier that ingests sequences of embeddings from the foundation model, and predicts if the traffic slice is under an attack. It leverages transformer architecture to detect subtle patterns across transaction sequences. And it does this all in real time so we can block attacks before they hit businesses. This approach improved our detection rate for card-testing attacks on large users from 59% to 97% overnight. This has an instant impact for our large users. But the real power of the foundation model is that these same embeddings can be applied across other tasks, like disputes or authorizations. Perhaps even more fundamentally, it suggests that payments have semantic meaning. Just like words in a sentence, transactions possess complex sequential dependencies and latent feature interactions that simply can’t be captured by manual feature engineering. Turns out attention was all payments needed!
175
544
236
5,189
1,304,147
Après un voyage de 2 semaines avec @royalairmaroc retour sur Casablanca, surprise baggage non encore localisé. Bravo #RAM pour cette médiocrité
2
1
257
1400$ pour un A/R et 2 bagages d'une valeur de 1500$ perdus par @royalairmaroc! Ils me disent que mes bagages arriveront à destination initiale après mon retour, soit une semaine après. Déception totale, service client inexistant. Voyageurs, soyez avertis ! #Scandale #ServiceClient
8
516
Voici la réponse de dame @royalairmaroc : Nous reviendrons vers vous dès qu'il y aura une information de l'aéroport. Bien à vous ! Depuis sans aucune nouvelle !
128
Highly recommended
Attached is the first session of the school. It's packed, and this is just Session 1. We have six of them! Classes start on Monday. You won't sleep and will have to cut down on social media, but you'll learn production machine learning. Sign up here: ml.school
430
Yassine LASRI retweeted
New short course with @googlecloud, which Nikita Namjoshi and I teach: Understanding and Applying Text Embeddings with Vertex AI! Come learn about text embeddings, a key part of a GenAI developer’s toolkit. You'll learn to apply embeddings to classification, outlier detection, text clustering, and semantic search. We also show how to combine text generation and semantic search to build a question-answering system. Please come join us! deeplearning.ai/short-course… – at Palo Alto, CA
11
173
15
964
160,879
Yassine LASRI retweeted
We're counting down the last two days until our incredible journey into innovation and technology begins at @GITEXAfrica We can't wait for you to join us at stand 10-C10 in Hall 10 and discover how we assist telcos and government regulators solve complex business problems.
1
1
223
Yassine LASRI retweeted
Come visit us during the @GITEXAfrica on Hall 10 Stand 10-C10 and discover how we use @Elastic technology to assist telecommunications companies and government regulators solve complex business problems. #regtech #analytics #ml #ai #bigdata #bigdataanalytics #elasticsearch #gitex
2
2
259
Yassine LASRI retweeted
Synaptique is looking forward to seeing you in #Marrakech for the @GITEXAfrica 2023 from May 31 - June 2. On the agenda, 3 days to explore the latest trends and technologies driving digital transformation across #Africa. Meet us on booth 10-C10
1
1
98
Yassine LASRI retweeted
My @NetflixEng colleague @amernetflix wrote up a nice high-level overview of the "noisy neighbors" problems one often encounters in large-scale container-based multi-tenant deployments. medium.com/@amerather_9719/n…
4
80
2
368
42,607
Yassine LASRI retweeted
Here’s how to enrich search analytics in Elasticsearch using your ingest pipeline: ✅ Resolve IP Address location ✅ Generate vector using trained ML model ✅ Enrich data using external data sources Learn more → go.es.io/3hrEjuJ
9
6
3,039
Yassine LASRI retweeted
Elastic has been recognized as a Leader in the 2022 @Gartner_inc Magic Quadrant™ for Insight Engines. Learn more about how Elastic gives developers extensive tools for building flexible, ML-powered search applications. Get the full report → go.es.io/3hzHyjK
18
32
3,541
Yassine LASRI retweeted
The broad set of new capabilities in Elastic 8.4 enables your organization to search, observe, and protect everything, across any environment. Read all about the enhanced Elastic Search Platform for fueling your search-powered solutions. Learn more → go.es.io/3QNrKGt
2
23
1
39
Notre rapport sur l'industrie du mobile money célèbre ses 10 ans et est maintenant disponible en français dans sa totalité! Téléchargez-le ici: https://nitter.gq/t.co/rXhvaBIgXC
3
4
Yassine LASRI retweeted
Most people overestimate what they can do in one year and underestimate what they can do in ten years – Bill Gates
116
5
683
#MobileMoney grew from a nascent offering in a few countries to a global financial service connecting hundreds of millions to formal financial services in the past decade. Read our 10th State of the Industry Report to learn more: https://nitter.gq/t.co/rXhvaBIgXC #SOTIR22
1
5
3
20
Yassine LASRI retweeted
Mobile Money Metrics are typically used to measure how well a business is performing against its goals. Discover More about Mobile Money Metrics in the link below: → lnkd.in/dTzgnYcY #money #mobilemoney #mobile #bigdata #analytics #data #telecoms #telcos #kpi
1
2
Yassine LASRI retweeted
Elastic 8.2 has arrived! This release enables you to build seamless search experiences, accelerate troubleshooting of cloud-native services, and enhance your security analysts’ investigative experience. Learn more → go.es.io/3y9xVOs
8
41
3
64