PAPIs Latam 2019: Full Schedule

09:00 GMT-03

Opening remarks

Speakers

Dan Nichol

Program Chair, AnalyticsFC

I build football data analytics tools for better recruitment at professional teams. Previously, I worked a postdoc studying the evolution of drug resistance in cancer.

Ana Paula Appel

Researcher and Data Scientist, IBM Research

Ana Paula is a Research Staff Member in IBM Research - Brazil, currently work with large amount of data to do Science WITH Data and Science OF Data at IBM Research Brazil. My technical interesting are in data mining and machine learning area specially in graph mining techniques for... Read More →

Louis Dorard

General Chair, PAPIs

Louis Dorard is the author of Bootstrapping Machine Learning, of the Machine Learning Canvas, General Chair of PAPIs.io (international conferences on ML applications and APIs), and Adjunct Teaching Fellow at UCL School of Management.As an independent consultant and Machine Learning... Read More →

Tuesday June 25, 2019 09:00 - 09:20 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

09:30 GMT-03

Reinforcement Learning for devs: toolkits, and application to autonomous race cars

Reinforcement Learning is a Machine Learning technique particularly suitable for complex, unpredictable, environments that can be simulated and where building a prior dataset would either be infeasible or prohibitively expensive. In this session, we'll introduce the main concepts of RL and pre-packaged RL toolkits available in Amazon SageMaker. We'll also talk about AWS Deep Racer, a fully autonomous 1/18th scale race car driven by RL.

Speakers

Julien Simon

AI Evangelist, AWS

As the Artificial Intelligence & Machine Learning Evangelist for EMEA, Julien focuses on helping developers and enterprises bring their ideas to life. He's also actively blogging at https://medium.com/@julsimon. Prior to joining AWS, Julien served for 10 years as CTO/VP Engineering... Read More →

Tuesday June 25, 2019 09:30 - 09:50 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

Keynote

10:00 GMT-03

Lessons Learned from Building a Credit Card Fraud Model

This talk covers two ML problems Stripe faced (and solved!) in building its credit card fraud detection system:

- Addressing class imbalance in training data: legitimate transactions greatly outnumber fraudulent ones, and models trained on imbalanced data often have low precision
- Performing counterfactual evaluation to measure performance in production and obtain unbiased training data when the ML system itself is constantly changing outcomes (if we block a transaction we think is fraudulent, we don’t actually know if we got it right—we have no way to observe the transaction’s ultimate outcome)

Speakers

Leela Senthil Nathan

Software Engineer, Stripe

Leela is a software engineer at Stripe, a technology company that builds economic infrastructure for the internet. She has spent two years working on various improvements to Stripe's transaction fraud models, including feature engineering, improvements to the training data generation... Read More →

Tuesday June 25, 2019 10:00 - 10:20 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

Keynote

11:00 GMT-03

Invoice Payment to Optimize Cash Collection

Predicting invoice payment is valuable in multiple industries and supports decision-making processes in most financial workflows. This work presents a prototype developed as a solution devised during a partnership with a multinational bank to support collectors in predicting invoices payment. The proposed prototype reached up to 83.5% of accuracy, which improved the prioritization of customers and supported the daily work of collectors and we expect to support researchers dealing with the problem of invoice payment prediction to get insights and examples of how to tackle issues present in real data

Speakers

Ana Paula Appel

Researcher and Data Scientist, IBM Research

Tuesday June 25, 2019 11:00 - 11:20 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

ML in the Wild

11:30 GMT-03

Where are the gains: Should I use A/B Tests for forecasting?

In this talk, we will analyze some theoretical aspects related to the results of an A/B test, in scenarios where Machine Learning models are involved. We will discuss the risks of taking results of an A/B test, or other purely comparative methodology, as a prediction of future gains. We will review the main practical and theoretical differences between A/B test and forecasting, when conducting ML related projects. This discussion targets a mixed audience, embracing people from technical and business areas. Also, it will be illustrated with real world examples taken from actual A/B tests.

Speakers

Eder Martins

Data scientist, SEEK

Eder Martins is a senior Machine Learning engineer passionate about solving problems. Currently, he is working on ML projects related to search and recommendation for online recruitment markets at Seek AI GDP team. MSc in Computer Science at UFMG, his interests include: Recommendation... Read More →

Tuesday June 25, 2019 11:30 - 11:50 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

ML in the Wild

12:00 GMT-03

I know what you did last session: clustering users with ML

Understanding user behaviour is one of the main challenges of any company. In this presentation, we’ll talk about how we explored users’ activity to find common patterns among them in three important moments of the customer journey. Through our experience with this project, we hope you’ll learn a bit about how to select the right data to analyze, how to find out which data actually matters, what are the challenges facing unsupervised learning algorithms in real life, how to check if your results make sense, and how you can use the power of storytelling to communicate results.

Speakers

Adauto Braz

Data Scientist, Stoodi

Adauto Braz - graduated at Aeronautics Institute of Technology (ITA) as a computer engineering in 2018 - is a Data Scientist, currently working at the Data team at Stoodi - an education startup. At Stoodi, he’s worked with Data Engineering, Data Analysis, Data Culture and NLP.

Tuesday June 25, 2019 12:00 - 12:20 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

ML in the Wild

14:00 GMT-03

Using Machine Learning to recommend jobs in User Cold Start

Industry still struggles to design recommenders able to provide useful recommendations for users who have not interacted with any item (User Cold Start scenario). In this talk, we will show how to connect new candidates to their desired jobs in one of the main online employment marketplaces in Latin America. The solution involves, first, modeling user and item profiles based on content features and then using Machine Learning to predict interactions between them. A / B testing evidenced the proposed solution is able to enhance user engagement.

Speakers

Andryw Marques

Data Scientist, SEEK

Andryw Marques is a Data Scientist at Seek-AI GDP team, currently working with projects related to Search and Recommendation for online employment. He has experience in fields such as Machine Learning, Information Retrieval and Data Analysis. Andryw received his Master and Bachelor's... Read More →

Tuesday June 25, 2019 14:00 - 14:20 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

ML in the Wild

14:30 GMT-03

A Clinical Application of Deep Learning for NLP with Word-Embeddings

Clinical coding represents the transposition of clinical findings and diagnostics into codes contained in the International Classification of Diseases (ICD). To perform such task, hospitals assign the role of “clinical coder” to the person responsible for reading the whole clinical documentation and assigning the ICD codes accordingly. This research aims to present a deep learning framework for natural language processing (NLP) capable of automating this task by interpreting (and further classifing) the text contained within these documents, by using "self-taught" word embeddings (learned from the database itself) as input.

Speakers

Arnon Santos

Data Scientist, Junto Seguros

MSc. (PUC-PR) in Computing for Health Sciences, researcher in IA and Machine Learning and member of the Brazilian Computing Society. My main research is based on the usage of Deep Learning applied to natural language processing, where I seek to expand the current state-of-art of natural... Read More →

Tuesday June 25, 2019 14:30 - 14:50 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

ML in the Wild

15:00 GMT-03

Time Series Forecasting for Cloud Resources Provisioning

Forecasting cloud resources consumption is one of the main challenges faced by IT operations teams in capacity planning tasks. Those teams often should make predictions for periods with no corresponding previous data whose cloud transactions activity changes, for instance, on a holiday. In this talk, it is discussed an approach which is able to model this future behavior with synthesized data. The procedure uses small samples to synthesize data to fit a time series forecasting model which applies interpretable parameters to be adjusted by domain knowledge analysts.

Speakers

Leonardo Neri

AI Manager, Accenture

Leonardo has a Ph.D. in Signal Processing and Pattern Recognition. His academic research is on Speech Processing field, specialized on Speaker Diarization.He is currently working at Accenture with data science, optimization problems, natural language processing, and machine learning... Read More →

Tuesday June 25, 2019 15:00 - 15:20 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

ML in the Wild

16:00 GMT-03

How Big Data Powers Ambev’s Sales Machine - sponsored by Big Data

Ambev visits hundreds of thousands of points of sale in Brazil every week, and needs to decide the optimal assortment and pricing for each sales call. In partnership with Brazilian AI leader Big Data, its sales process was transformed into a machine-learning powered sales machine, using thousands of variables to deliver actionable insights for sales reps, bringing significant sales growth. This session will present the real-life experience of one of the world’s largest CPG companies as it deploys the knowledge frontier at scale

Speakers

Gustavo Ioschpe

Founder & CEO, Big Data

B.S. Wharton School, Management; B.A. University of Pennsylvania, Political Science. M.A. Economics, Yale University. Founder and CEO @ Big Data, Brazil's leading applied IA company. Member of the board at Iochpe-Maxion S.A., Instituto Ayrton Senna and Conselho de Gestão da Educação... Read More →

Tuesday June 25, 2019 16:00 - 16:20 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

Keynote

16:30 GMT-03

How Machine Learning is Transforming the Way iFood Runs its Logistic - sponsored by Movile

Food delivery logistics pushes the boundaries of traditional logistics with new characteristics: very short times to delivery (<40mn), the use of a cloud of drivers, and deliveries route optimization taking into account stochastic future events. IFood is reinventing the way it runs logistics with new fields of machine learning applications including the repositioning of drivers to meet demand, dynamic pricing to manage driver supply and the estimation of logistic delivery times. We will analyse in detail how models of machine learning were built and deployed at scale to estimate delivery time of arrival, and kitchen preparation times, in order to improve final customer experience and reduce logistic costs.

Speakers

Arnaud Seydoux

Logistic IA manager, iFood

IA and modelling enthousiast, I graduated in evolutionary genetics, currently leading logistic IA team at iFood.

Tuesday June 25, 2019 16:30 - 16:50 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

Keynote

17:00 GMT-03

Panel: Diversity in ML

ML models are known to learn from the data they are submitted. However, some problems in the models can only be accounted for when they fail to express the "impartial" perspective they were supposed to have. In this sense, to get the most benefit from advances in AI and machine learning, culture, and diversity in the teams is a critical aspect to avoid data bias in many senses. Without diverse teams, the data and the learned models may drive automation to a bias that only deepens the gap.

Speakers

Esther Colombini

AI Researcher, Advanced Institute for Artificial Intelligence

Professor of Robotics and Artificial Intelligence @ University of Campinas (Unicamp). Ph.D. and an MSc degree in Computer Engineering @ Technological Institute of Aeronautics (ITA).Bachelor on Computer Science @ Universidade Federal da Paraíba (UFPB), partially executed at Unive... Read More →

Larissa Lautert

Data Scientist, Linx Impulse

Computer scientist working on data science. Experience in predictive models, recommender systems, ETL, A/B test evaluation and data analysis. Passionated by translating large data sets into business insights and answering questions with data. Analytical and thorough profile.

Ana Paula Appel

Researcher and Data Scientist, IBM Research

Ricardo Herrmann

Artificial Intelligence Engineer, Olivia AI

Ricardo Herrmann is a Computer Scientist working as an Artificial Intelligence Engineer at Olivia AI. He has 20 years of experience in the industry, having previously worked as Data Scientist at Dell EMC Consulting and as a Research Software Engineer at IBM Research Brazil, among... Read More →

Tuesday June 25, 2019 17:00 - 17:30 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

Panel

09:00 GMT-03

Ludwig, a Code-Free Deep Learning Toolbox

The talk will introduce Ludwig, a deep learning toolbox that allows to train models and to use them for prediction without the need to write code. It is unique in its ability to help make deep learning easier to understand for non-experts and enable faster model improvement iteration cycles for experienced machine learning developers and researchers alike. By using Ludwig, experts and researchers can simplify the prototyping process and streamline data processing so that they can focus on developing deep learning architectures.

Speakers

Piero Molino

Senior ML / NLP Research Scientist, Uber AI Labs

Wednesday June 26, 2019 09:00 - 09:20 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

Keynote

09:30 GMT-03

DataOps architecture for Machine Learning - brought to you by everis

As DataOps practices are changing the way that corporations deal with their data, there is a relevant opportunity to reduce the risk and time-to-market inherent to every Machine Learning project. However, are you ready to adopt those practices? In this presentation, you will get to know Noronha: a new open source, user-friendly framework that aims to host your projects inside a portable, ready-to-use DataOps architecture, thus helping you benefit from the most trending DataOps practices without having to change much of your usual work behavior.

Speakers

Carlos Porto Filho

Data Scientist, Everis

Data scientist at Everis Brazil. I have a bachelor’s degree in Biomedical Informatics and a master’s degree in Bioengineering. I have worked in machine learning solutions in health, banking, retail and telecom.

Gustavo Castilhos

Data Architect, everis

I am a Data Engineer and Architect at everis, where I assist our Machine Learning team by designing both batch and real-time pipelines that allow them to extract value of their Big Data assets with the most efficiency and automation. My latest works are focused in the engineering... Read More →

Wednesday June 26, 2019 09:30 - 09:50 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

Keynote

10:00 GMT-03

AI culture and semi-autonomous review approval - sponsored by Dafiti

Dafiti Group, Latin America's biggest fashion e-commerce is on its way to become an AI-first enterprise. During this transformation it is essential to understand how to find AI opportunities throughout all Dafiti's areas and how to map such opportunities to OKRs (Objective and Key results). Under these conditions, we present in this talk as a concrete example how we created a system which evaluates - approves and rejects - user’s product reviews automatically, along with it provides a tool for easier data curation by humans, sample analysis and monitoring.

Speakers

Ricardo Savii

Data Scientist, Dafiti Group

Wednesday June 26, 2019 10:00 - 10:20 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

Keynote

11:00 GMT-03

BERT: Multi-language approach for Q&A and NLP Applications

The advent of deep pre-trained models represents a big change in the way natural language processing is implemented. We present here a state-of-the-art technique called Bidirectional Encoder Representation of Transformers (BERT). It is as an approach to tackle the Question & Answers task as well as others NLP problems in a multi-language environment. We show the benefits and drawbacks of using such solution in real world applications.

Speakers

Horst Rosa Erdmann

Lead Data Scientist, Everis

Computer Engineer with Master Degree in AI on FEI UniversityWith large experience in Big Data and Data Science projects.Worked in several areas in AI such as Computer Vision, Natural Language Processing, Time Series, Signal Processing, Fraud Detection among others.Joined one of the... Read More →

Wednesday June 26, 2019 11:00 - 11:20 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

New techniques

11:30 GMT-03

Multitask convolutional neural networks: saving GPU time

Convolutional neural networks are great for image classification, there's no doubt about it. The problem is: each one you add to your pipeline increase processing time and require more GPU memory, which can get quite costly. We'll discuss how to deal with that by training convolutional neural networks with multiple classifiers sharing the convolutional layers. We'll evaluate the results in a dataset of apparel products, discussing impact in accuracy and on the quality of the embedding.

Speakers

Paulo Eduardo Sampaio

Data science specialist, McKinsey & Company

Paulo is a civil engineer with a MBA and a MSc in statistics and operations research. He's been working with applied machine learning since 2013 and from 2015 onwards he specialised in computer vision applications. He's currently working for McKinsey as a data science specialist in... Read More →

Wednesday June 26, 2019 11:30 - 11:50 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

New techniques

12:00 GMT-03

AI informs, humans choose -- or do they?

The first expert systems were aimed at informing people so they made better decisions and depended less on hard-to-find-harder-to-pay human specialists. Now, however, with the increasing refinement of techniques, hidden layers, and ubiquity of complex systems, results are often applied as they come. Humans are not good at making decisions under risk, and in Machine Learning applications they often don't even know they're taking a rough probability as an inexorable result. So, how may we design systems that help humans make choices with healthy doses of skepticism?

Speakers

Bianca Ximenes

Independent consultant / Doctoral Student, Tecnora / UFPE

Bianca is a Google Expert on Product Strategy, Currently, as a Doctoral Student in Computer Science at UFPE, her work approaches human-machine boundaries and ethical repercussions of AI and ML on digital products.She enjoys reading, writing, and talking about human-AI coexistence... Read More →

Wednesday June 26, 2019 12:00 - 12:20 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

New techniques

14:00 GMT-03

Game Theory and Model Interpretability

Many machine learning models are (in)famous for being black boxes. However, model explainability is a hot topic and new techniques have been brought to life in order to overcome that. This talk will walk through how a 60+ year-old game theory concept has been adapted so that we can interpret machine learning models regardless of their inner structure and complexity — yes, neural nets and ensembles included. Most importantly, we'll talk about how you can apply this technique today in order to build better models and come up with more useful data products.

Speakers

Gabriel Cypriano Saca

Data Scientist, Grupo ZAP

Gabriel is currently a Data Scientist at Grupo ZAP and a Data Science Instructor at Tera. He has previously worked as a Data Scientist at Creditas and at K2 Data Science. Gabriel discovered the world of Machine Learning while co-founding Songwich, a music recommendation startup.

Wednesday June 26, 2019 14:00 - 14:20 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

New techniques

14:30 GMT-03

Trading strategies using deep reinforcement learning

A trader is defined as an entity that buy and sells financial instruments, while a trading strategy is a fixed plan that is designed to achieve a profitable return. The goal for the work presented here is to train a reinforcement learning agent, which is acting as the trader, to learn how to propose a good trading strategy. The agent considers not only the current information of the market but also some information of the news around the analyzed financial instrument.

Speakers

Suraj Shinde

Director - AI Digital Lab, EVERIS MEXICO S DE RL DE CV

Suraj Shinde is Director of AI Digital Lab at everis, which is focused on research and development of artificial intelligence related technologies and solutions including developing proofs of concept, intellectual property and assets. He is also Head of everis AI in Mexico which is... Read More →

Cristyan Rufino Gil Morales

AI Research Lab Leader, everis

Self-taught person, goal-oriented, focused on AI solutions and passionate for hard challenges.Experience in the use of Reinforcement Learning algorithms and its combination with robotics. Wide experience using deep learning and machine learning models.

Wednesday June 26, 2019 14:30 - 14:50 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

New techniques

15:00 GMT-03

Attention-Based Neural Networks for Relational Data

Attention-based neural networks have recently achieved excellent performance on machine translation tasks. A 2017 NIPS paper from Google , “Attention Is All You Need”, demonstrated that attention mechanisms have the potential to replace recurrent neural networks, which are known to be computationally expensive and at times unstable. We provide a detailed example of how the same ideas can be applied to relational databases, by using attention layers which learn to condense the information present in a collection of related data tables, without the need for recurrent layers or hand-crafted features.

Speakers

Scott Brownlie

Senior Data Scientist, TOTVS Labs

Senior Data Scientist at TOTVS Labs. Holds a BSc in Mathematics from the University of Glasgow and an MSc in Mathematics from the University of Cambridge.

Wednesday June 26, 2019 15:00 - 15:20 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

New techniques

16:00 GMT-03

How Grupo ZAP is using data to empower real estate buyers, sellers, and renters in Brazil - sponsored by Grupo ZAP

Real estate in Brazil is one of the industries that lacks transparency the most. Such structure triggers a lot of dissatisfaction among all stakeholders. Grupo ZAP uses data to predict market prices, demand, and other real estate indexes that drive value, helping consumers to buy, sell, and rent faster, better, and in a more convenient way.

Speakers

Lucas Vargas

CEO, Grupo ZAP

Wednesday June 26, 2019 16:00 - 16:20 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

Keynote

16:30 GMT-03

Machine Learning for Natural Resources

In order to identify new exploration targets to test potential zones of gold mineralization, geologists gather different sources of data such as drill hole observations and measurements to justify their decisions. Such activity is complex and requires expert tacit knowledge specific to each mining project, meaning that the location's underlying geology is unique. With the aim of aiding this decision making process, we present a machine learning methodology to predict the level of gold mineralization based on the surrounding geological information. In particular, we propose the use of Deep Learning models, 3D Convolution Neural Networks, to take advantage of the spatial nature of the data. Through experiments with data from a real mine, we show that our proposed model overcomes the baseline model for all the metrics used, providing a much more accurate feature for the geologists in their investigations.

Speakers

Bianca Zadrozny

Research Manager, IBM

Bianca Zadrozny is a research manager at IBM Research Brazil, leading the Natural Resources Analytics group. The group's mission is to conduct research projects in knowledge-augmented machine learning for decision making in the areas of oil&gas and mining, with a great focus in developing... Read More →

Wednesday June 26, 2019 16:30 - 16:50 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

Keynote

17:00 GMT-03

Panel: How much should we care about model interpretability?

Model interpretability is getting more and more attention both from ML practitioners and the community at large. With the GDPR and LGPD coming into force and discussions about privacy, fairness, and ethics becoming more relevant than ever, while models are getting more and more powerful with increasing complexity and opacity, how much we, as professionals and as a community, should care about model interpretability?

Speakers

Renato Vicente

Chief Scientist, Serasa Experian

A physicist interested in Machine Learning, Information Theory and Complex Systems since 1992.

Ivan Marin

Data Scientist/System Architect, Daitan Group

Physicist by training, with Masters in Applied Physics and PhD in Engineering, Ivan has experience outside the academia with applied data science, big data and architecture in different contexts, like Telcos, Healthcare, Agribusiness and other areas.

Sandor Caetano

Chief Data Scientist, IFood / Movile

Sandor Tucakov Caetano, is an Economist (FEA-USP). Worked as a Data Scientist before it was fashionable and since 2006 built marketing, pricing and sales models for big consumer goods and food companies. Worked as the head of Data Science for a leading Fintech company and today is... Read More →

Wednesday June 26, 2019 17:00 - 17:30 GMT-03
Room 8 Av. Rebouças, 3970 - Pinheiros, São Paulo - SP, 05402-600, Brazil

Panel

17:30 GMT-03

Closing remarks

Speakers