Full Machine Learning Week 7-Track Agenda 2023 – Detailed Session Descriptions

Machine Learning Week Founder Eric Siegel will welcome you to the event and kick things off.

Machine Learning Week Founder Eric Siegel will welcome you to the event and kick things off.

Machine Learning Week Founder Eric Siegel will welcome you to the event and kick things off.

Machine Learning Week Founder Eric Siegel will welcome you to the event and kick things off.

Machine models are the most powerful predictors, but they are often black-box models and incorporate uncertainty by nature. Quantifying the ML uncertainty is critical for adding confidence to ML adoptions. Based on reliable uncertainty estimators,  risks of ML underperformance can also be quantified and transferred through insurance solutions. We will demonstrate this application through real case studies of collaborations between ML providers and the insurance industry.

International students, DACA, immigrants, and other new Americans represent 13% of the US population and are growing. However, a lack of existing credit history prevents this population from accessing affordable loans. In this case study, we’ll share how MPOWER Financing – a large provider of financing to international and DACA students – uses an innovative, forward-looking credit model based on alternative data and new modeling techniques to provide affordable education financing to people with no or limited credit history.

In this session, we will report on major learnings and takeaways for translating AI models into clinical practice. Mayo Clinic has created a systematic assessment of AI models using multidisciplinary subject matter expertise for projects at all phases of development (from ideation to implementation and monitoring) to assess whether AI is the right solution, advise on next steps for regulatory considerations, make recommendations for electronic health record implementation, facilitate the clinical endorsement process, etc.  

Machine learning (ML) is a rapidly growing field that opens a lot of opportunities for transformation workflows for many users. This presentation will introduce how to enhance user productivity using Machine Learning and replace repetitive manual actions with intellectual ML suggestions. This talk will cover the following:

• Developing a universal ML workflow to process any type of users data (image and text)
• Instrumentation and feedback collection that will help drive future improvements by using incremental learning techniques
• Best practices in applying ML for optimizing the user experience working with Autodesk product

Machine Learning (ML) models are quickly becoming ubiquitous and widely applied in banking. However, the use for high risk applications such as credit underwriting demands higher requirements in terms of model explainability and testing beyond performance evaluation. In this talk, I am going to share how we approach model interpretability without the potential pitfall of post-hoc explainers by employing inherently interpretable machine learning. Beyond interpretability, comprehensive testing to ensure model robustness and resilience are required for high risk applications. I'm going to discuss our approach illustrated by a tool that we recently released, PiML (Python Interpretable Machine Learning), an integrated environment to develop and validate machine learning models for high risk applications.

Despite the rapid evolution of AI, projects still fail at a disappointingly high rate. In the past, capturing data at scale and building models was the challenge, but today we're confronted with the issue of making AI more robust while avoiding the risk of unintended consequences. While the tools are new, many challenges remain the same. In this talk, I will share by way of real-world examples improving business processes at University Hospitals, a tier 1 trauma hospital: * How to build the business case for an AI project (and get buy-in) * Navigating AI project management to prevent failure * How to mitigate the risks of unintended consequences from using A

The massive dimensionality of genomic data combined with the lack of knowledge of gene function have been major barriers to the use of Artificial Intelligence (AI) in genomics. The number of population sequencing initiatives is rapidly increasing. As these large genomic datasets grow and become more widely available, the prospect of using artificial intelligence to gain insights from this data becomes more tangible and the use of such technology becomes even more important. In this session we review the current state and future directions of using AI with genomic data.

Rexer Analytics began surveying data scientists in 2007. This year's Data Science Survey is a collaboration with "Predictive Analytics" author and conference series founder Eric Siegel. In this session, Karl and Eric will present preliminary results from this year's survey. Topics will include algorithm choices, data science job satisfaction trends, deep learning, model deployment, and deployment challenges.

Often data science teams, particularly in non-academic health systems where resources are scarce, will face difficulty in securing the green light for internal predictive modeling efforts. Along with technical requirements, equally important for realizing value are the strategies in presenting the business case for internal development to executive decision-makers. This case study will cover the process of advocating for data science teams in build vs. buy discussions through deploying a custom and adjustable Target Length of Stay model within clinical workflows in partnership with end users.

At Northwestern we have developed a system that is built to consume and capitalize on IoT infrastructure by ingesting device data and employing modern machine learning approaches to infer the status of various components of the IoT system. It is built on several open source components with state-of-the-art artificial intelligence. We will discuss its distinguishing features of being Kubernetes native and by employing our work that enables features to be specified through a flexible logic which is propagated throughout the architectural components. We will discuss select use cases implemented in the platform and the underlying benefits. The audience will learn how to build or use a streaming solution based on Kubernetes and open source components.

Reinforcement Learning (RL) agents proved to be a force to be reckoned with in many complex games like Chess and Go. Financial firms are leveraging the power of RL, given it potential to automate all the steps involved in algorithmic trading. However, it is quite challenging to understand and interpret a RL based models.

This talk focuses on an approach to understand and interpret Reinforcement Learning (RL) based trading strategies. We first briefly introduce the concept of reinforcement learning in the context of algorithmic trading, followed by demonstration of an RL- interpretability infrastructure. We then discuss possible derived outcomes of using this infrastructure when applied to trading a market instrument.

Our care providers have challenges, but those challenges rarely arrive looking like data science questions.  How do we use the skills of data scientists to deliver what providers really need?

How to Attract and Recruit Analytics Talent from Top Universities (especially if you're not a tech giant)Competition for top new analytics talent is fierce. While tech and other corporate giants are indeed vacuuming up new grads from top schools, not all great students can or want to go that route. The challenge is to put yourself in a position to attract and land them. It can be done, even if you're not a well-known brand. In this session, you will learn what works from a leader of the analytics program ranked #2 in the world by QS for the past three years.   Even if you are from a giant firm, you're still competing for talent. You will come away with ideas to help you gain an advantage!

Enjoy some machine learning laughs with Evan Wimpey, a predictive analytics comedian (and we're not just talking about his coding skills). No data topic is off-limits, so come enjoy some of the funniest jokes ever told at a machine learning conference.*

* Note the baseline.

Enjoy some machine learning laughs with Evan Wimpey, a predictive analytics comedian (and we're not just talking about his coding skills). No data topic is off-limits, so come enjoy some of the funniest jokes ever told at a machine learning conference.*

* Note the baseline.

Conference Chair Chris Franciskovich will welcome you to day 2

Conference Chair Steven Ramirez will welcome you to day 2

Rexer Analytics began surveying data scientists in 2007. This year's Data Science Survey is a collaboration with "Predictive Analytics" author and conference series founder Eric Siegel. In this session, Dr. Rexer will present preliminary results from this year's survey. Topics will include algorithm choices, data science job satisfaction trends, deep learning, model deployment, and deployment challenges.

In the life insurance industry, underwriting is an important process to assess the insurability of an individual. Traditional underwriting is often lengthy and expensive due to the requirement of blood and urine tests. The accelerated underwriting process replaces these lab tests with data and machine learning algorithms, thereby saving both time and money for the policyholders. The process made life insurance purchases easier and faster, thereby allowing the industry to serve a wider demographics and improve social good. This talk seeks to provide an overview of the AUW, emerging data sources, best practices, challenges, and regulatory implications.

This case study presents how Hospital Israelita Albert Einstein, leading Latam healthcare organization, utilized advanced analytics to determine the optimal layout for their new Cancer Center hospital. In 2021, as part of their expansion plan, they have decided to build a new Cancer Center in São Paulo, migrating their oncological operation to the new site.

The project for the new hospital was based on three pillars: Clinical Safety, Architecture and Operational Efficiency. The latter was the focus for the analytics’ team effort, using process mining, forecasting the future demand and optimization to generate the ideal layout to minimize movement flows.

The approach taken was based on 3 workstreams:

• Use of process mining  to generate clinical pathways (CP) for all oncology patients in the current operation
• Clustering algorithms to classify the patients considering the different medical specialties used in their CPs
• An optimization model was developed, using mathematical models and metaheuristics, to receive the future demand for each CP along with resources’ requirements and generate the optimal layout suggestion

We captured impressive results: >15% reductions in movements for patients, visitors and materials and >8% for health staff, generating cost savings of approximately 3% of total operating cost.

Within the Human Resources industry there are still many challenges ahead, for example, keyword-based searches or manual resume screening could contribute to omitting valuable opportunities or lengthy hiring procedures. For this reason, we would like to present a novel Recommendation System in the field of HR, which includes a series of Natural Language Processing techniques and Deep Learning models, that allow us to achieve a fully automated process that will propose semantically related and explainable suggestions to job seekers and companies in real-time. The system architecture combines several NLP techniques, such as named entity recognition, text classification, and entity relationships based on HR data, to extract and process more than 50 different characteristics of job postings and resumes, such as occupations, skills, education, etc., with various granularity levels and multiple languages.

Retail shops need credit to purchase inventory. Credit dependence is more prominent with small retailers (mom & pop stores).  In India, most of these stores don't use digital transactions or worse, don't even have bank accounts. Due to this, they can't get loans from banks as no credit score is present. In absence of formal credit, retailers depend on distributors (businesses who supply retailers inventory) to extend them line of credit (informal credit). Most of this credit offering is based on historical relations between retailer and distributor with no scientific premise. This credit is based on personal relationships without taking into consideration the actual worth of the shop -- therefore, there are high risks of loan default.

With data access of nearly 7.5MN retailers on real time basis we have enough retail transaction level data & hyper local data to train models that can predict credit worthiness of a store without need for any banking information . In this talk, I intend to explain how we use data along with the ML models to build Credit Score for Retailers. A major financial lender use this score to extend credit to retailer achieving lowest delinquencies in the industry

This talk provides a 3-dimensional framework for understanding professional competence and career progression. The framework covers primary skills (analytics), complementary skills (oft overlooked "soft skills"), and impact (how one's work advances the mission). Once defined, the speaker provides actionable guidance on how to develop a roadmap to achieve raises, bonuses, and promotions via a one-year plan. The content is optimized for early career professionals (<5 year’s experience) in need of guidance on how to move up and leaders of early career professionals in need of a framework to help cultivate their talent.

We illustrate here a Living Digital Twin of a fleet of Electric Vehicles that gives actionable predictions of battery degradation over time. Since each vehicle takes a different route and has different charging and discharging cycles over its lifetime, the battery degradation for each vehicle will be different. We use a scalable predictive modeling framework deployed across a distributed computing architecture in the cloud to make individualized predictions for each EV battery in the fleet to reflect the degraded performance accurately. The neural network battery model is calibrated using the parameters that had the most significant impact on the output voltage through the Unscented Kalman Filter (UKF) method, which is a Bayesian technique for parameter estimation of non-linear system behavior. We simulated in-production real-world operations by having the vehicles “drive” the routes using synthetic datasets and showed how the calibrated model provide more accurate estimates of battery degradation. Using a Living Digital Twin to calculate the remaining range and battery State of Health addresses problems of range anxiety in the EV automotive industry and can drive the value of EVs in the market.

ML’s great strength is that example cases are all you need to create a predictive model. The predictions work as long as the underlying process is not tampered with. But clients usually seek more: they yearn to understand the "data-generating machinery” in order to improve the outcome that the model predicts. Yet, this is dangerous without additional external information, including the direction of influence between variables. This talk illustrates how to achieve “peak interpretability” by using influence diagrams to model causal relationships, avoid mistaking correlation for causation, and quantify how outcomes will change when we manipulate key values.

ML’s great strength is that example cases are all you need to create a predictive model. The predictions work as long as the underlying process is not tampered with. But clients usually seek more: they yearn to understand the "data-generating machinery” in order to improve the outcome that the model predicts. Yet, this is dangerous without additional external information, including the direction of influence between variables. This talk illustrates how to achieve “peak interpretability” by using influence diagrams to model causal relationships, avoid mistaking correlation for causation, and quantify how outcomes will change when we manipulate key values.

ML’s great strength is that example cases are all you need to create a predictive model. The predictions work as long as the underlying process is not tampered with. But clients usually seek more: they yearn to understand the "data-generating machinery” in order to improve the outcome that the model predicts. Yet, this is dangerous without additional external information, including the direction of influence between variables. This talk illustrates how to achieve “peak interpretability” by using influence diagrams to model causal relationships, avoid mistaking correlation for causation, and quantify how outcomes will change when we manipulate key values.

Model explainability or interpretability is often demanded, posed as a requirement. But not always. Under what circumstances is it pragmatically necessary (or even legally required) -- and, when it is called for, what exactly does it mean? Would it suffice to explain each model prediction by showing what differences in inputs would have changed the prediction? Or must the model be "understood" globally? Is understanding how the model derives its predictions sufficient, even without the why -- that is, without causal explanations for the correlations it encodes?

Join this expert panel to hear seasoned experts weigh in on these tough questions.

Model explainability or interpretability is often demanded, posed as a requirement. But not always. Under what circumstances is it pragmatically necessary (or even legally required) -- and, when it is called for, what exactly does it mean? Would it suffice to explain each model prediction by showing what differences in inputs would have changed the prediction? Or must the model be "understood" globally? Is understanding how the model derives its predictions sufficient, even without the why -- that is, without causal explanations for the correlations it encodes?

Join this expert panel to hear seasoned experts weigh in on these tough questions.

Hospital beds are an important, finite resource in the healthcare system, and effective utilization of this resource is key to both providing the highest standard of care as well as ensuring operational efficiency. However, the discharge process is often a bottle neck due to the necessity of completing time consuming discharge paperwork as well as the need to identify non-hospital healthcare facilities with capacity to transfer patients that need continuing care. In this case study, we use patient interaction data during the hospital stay to build a predictive model for patient discharge type (defined as both a binary variable and a multi-class classification problem) allowing discharge staff to preemptively begin these time consuming steps in the discharge process. This allows for earlier discharge of patients and more effective utilization of the hospitals resources. While this system has not been implemented at this time, we provide a framework to evaluate the economic value of a predictive model within this context.

Industrial manufacturing has gained largely from artificial intelligence (AI). Particularly, generative design has leveraged cloud computing and AI to allow fine-grained control over manufacturing processes, loads, and constraints. On one hand, AI-powered solutions simulate complex geometries with customizable material properties. On the other, AI has enabled control over processes without compromising design safety and with significant material savings. Additionally, the foundation models in computer vision and natural language have opened the doors to limitless possibilities.

Delivering a product that needs to meet global high-quality standards requires controlled design and manufacture, navigating the often confusing logistical supply chain, and customer-driven time-to-market deadlines. According to Forbes, unplanned downtime costs industrial manufacturers $50 billion a year. As manufacturing companies scale AI initiatives, the struggle remains to align their operational needs with AI capabilities.

The heavy reliance on limited or siloed data can lead to severe compromises on accuracy and the value of insights driven by AI systems. Moreover, with the increasing complexity of the product and strict quality-control standards, fast-moving production companies lag behind in being able to explain AI models. This inhibits the visibility into patterns that can lead to potential problems or failures that result in unplanned outages and low production efficiency.

Join a session with Ayush Patel, Co-founder at Twelvefold as he takes you through the AI-enabled generative design landscape and its challenges. Explore key strategies to maintain scalable yet reliable AI solutions while achieving significant ROI from manufacturing operations.

Predictive analytics can be applied to pricing car insurance to help insurance companies determine appropriate rates for individual policyholders.  By analyzing large datasets of historical insurance claims and policyholder data, predictive models can be developed to estimate the likelihood of future claims based on various factors such as age, gender, driving history, location, credit score, and type of vehicle.  More recently, insurers have also been able to use the rich datasets from telematics sensors.  This session will provide an overview of how predictive analytics can be applied to pricing car insurance to help insurance companies make more informed pricing decisions and reduce risk, ultimately leading to improved profitability and customer satisfaction.

Leading healthcare experts will discuss how they see data science in healthcare evolving in the near, mid and long term future.

Supply chains have been very lean for decades, and the challenges in the last few years have led more teams to focus on resiliency. With the quantity of data available across manufacturing, shipping, stocking, and inventory, teams can turn to analytics to improve the resiliency of supply chains. This talk will highlight how analytics has helped improve supplier reliability, decrease sourcing costs, and quantify the true risk of supply chain disruption.

In this presentation, Finicity lead data scientist Natesh Arunachalam will provide a brief primer to Open Banking. He'll cover why Open Banking is a crucial technology for FinTechs and Banks, why it provides more power to the digitally native consumer, data assets that Open Banking provides, and application of AI in Open Banking, including use cases in risk scoring, NLP, and computer vision.

Defining the problem to solve for, defining the desired outcomes, understanding the type of data involved, and preparing the data appropriately are the four most critical aspects of machine learning in the research space – which will be presented in this session.

Ask our “Rockstars” anything about predictive analytics! Curious about machine learning tips, tricks, best practices and more? This is your opportunity to hear advice directly from the experts. The hardest part of machine learning can be getting models deployed into production. Our panel has done it, and is willing to spill the tea on how to make it happen. You’ll want to stick around for this ultimate session of the conference.