Y-DATA Industry Projects 2021 

/Healthcare
Healthy.io
K-Health
Medisafe
/FinTech
Tipalti
/Medical Imaging
AInovIA
Body Vision Medical
Zebra Medical Vision
/Market Intelligence
Gong.io
Guesty
Roojoom
Similarweb
/Retail Tech
Supersmart
Weissbeerger
/Content Discovery
Taboola
/BioTech
Compugen
Evogene
/Maritime Intelligence
Windward
/Manufacturing
ICL
Nvidia
/Cyber
TRG
TSG

About

Industry Projects are one of the key elements of Y-DATA program, translating theoretical knowledge into practical skills. We believe that in order to achieve full understanding of the use and application of machine learning algorithms, Y-DATA students should experience, as part of their studies, work on a real-life, full-cycle industry project. This experience will provide invaluable hands-on experience working on real data and problems and overcoming realistic challenges.

Presented here is the list of industry projects available for the current course year and background information on the companies offering them, as well as the logistics for each project and company.

Read More

Industry Projects

AInovIA
AinovIA Labs is an innovation startup launchpad established by an experienced entrepreneurial team.  Through our work with partners and customers we identify market problems and gaps that can be addressed with AI. Our experienced AI teams develop relevant products & technologies that we commercialize or spin out as separate startups.

Project:
Sperm Detection and Classification

Our mission is to create a model for detecting and classifying human sperm from microscope videos. We have a unique team of specialized technicians that is tagging our videos using CVAT tool - a tool for image annotations that is widely used (Computer Vision Annotation Tool). We use a framework to implement this cycle. This framework can tag, train and evaluate vision models with a click of a button(s). There are many features that are missing out of the box. The goal in this project is to allow a full cycle of tag-learn-infer.

Body Vision Medical
Body Vision Medical was founded in 2014. Our mission is to provide an affordable and effective real-time intra-body navigation platform for early-stage cancer diagnostics and treatment. LungVision™ System is the the main product - a revolutionary platform, providing immediate access to early stage lesions that can still be cured

Project:
Preoperative and intraoperative CT registration

Body Vision's platform provides an affordable and effective real-time intra-body navigation platform for early-stage cancer diagnostics and treatment. It is a powerful imaging device providing the capabilities of planning target and pathway for real-time navigation during operative procedure. Our product is using image registration between pre-operative information that we extract from a CT image and intraoperative information. We want to solve this problem differently by using our recent development of ability to create intraoperative CT, and to solve the registration problem by registering the preoperative CT with the intraoperative CT.

Compugen
Compugen is a clinical-stage drug discovery and development company utilizing a broadly applicable, predictive computational discovery platform to identify novel drug targets and new biological pathways and develop therapeutics in the field of cancer immunotherapy.

Project:
Predicting protein levels from transcriptomic data in single cells

The quantification of the different proteins is of great scientific value, yet such information at the single-cell level is currently lacking. Measurement of mRNA transcripts levels in single cells is an emerging technology that is transforming biomedical research and drug discovery. Currently, most public datasets measure only mRNA levels, and only a minority measure also proteins. Training a model that would be able to predict protein abundance based on mRNA quantities is highly valuable for augmenting existing datasets.

Evogene
Evogene is a leading biotechnology company developing novel products for life science markets through the use of its unique products in the fields of small molecules, genes and microbials discovery (ChemPass, GeneRator and MicroBoost). These products enhance the life science products development in the fields of agriculture, human health and industrial implications.

Project:
Automatic representation of molecules as complicated features

Many prediction models dealing with small molecules' properties rely on the representation of large datasets of compounds and the generation of predictions based on expert-crafted features. During the last few years, various types of such features were generated, with powerful new approaches emerging allowing the processing of huge databases and the generation of complicated predictive models. In this project, you will develop a property prediction algorithm based on advanced molecular features.

Project:
Bacterial plasmid identification in metagenomic data

In this project, you will first review the current existing algorithms for plasmid identification in metagenomic data and then recommend either the incorporation of an existing algorithm, with or without its modification, the incorporation of several tools or creating a novel computational tool for plasmid identification in metagenomic as well as genomic data.

Gong.io
Gong.io uses artificial intelligence (AI) to analyze conversations, enabling sales teams to better understand the things that are going right – and wrong – in their sales calls. Gong automatically records, transcribes, and analyzes sales calls, helping sales reps to improve their pitch and win more deals. The company analyzes conversations from audio sources and web-conferencing platforms such as Cisco WebEx, GoToMeeting, and Zoom, and then links the results to CRM systems such as Salesforce.com. With Gong, real-life listening, pitching, probing, positioning, and closing skills are suddenly available for review and refinement.

Project:
Speeding up Transformer-based NLP models

In the past 2 years, NLP has had its "ImageNet moment". Novel attention and transformer-based Neural Network (NN) architectures smashed previous state-of-the-art results, and transfer well to many downstream tasks. But while transformers bring a huge performance boost, they have a significant computation toll. Hand-in-hand with the development of more accurate transformer-based models is the reduction of the computational requirements for inference, using techniques like distillation, quantization, pruning and more.  In this project, you will train NLP models based on BERT, ELECTRA, ROBERTA and other models using a GPU, and then experiment with various methods to reduce their complexity and run times on a CPU.

Guesty
Guesty is the ultimate property management platform for short-term and vacation rentals. Our end-to-end solution simplifies the complex operational needs that property managers face on a daily basis – from guest communication to task assignment to payment processing. With Guesty, property managers save time so they can focus on what matters most: growing their business.

Project:
Price Prediction For Short-Term Rental Properties

One of the biggest challenges for Property Managers is how to set a property's price correctly - so as to allow them to pick reservations and optimize their future revenue stream. Guesty is interested in helping our clients understand the market conditions better, and map their properties according to these conditions in terms of price. For this, we would like to develop a Machine Learning model that will predict the optimal price per property.

Healthy.io
Healthy.io is one of the first companies to successfully turn a smartphone into a clinical-grade medical device. It has products covering a wide range of functions, from early detection of chronic kidney disease, digital management and assessment of chronical wounds, smartphone-based detection and treatment of UTIs, home-testing kits streamlining prenatal care and more.

Project:
Evaluation of DL keypoint-matching approaches for structure from motion

Healthy.io is turning smartphone cameras into clinical devices by leveraging computer vision algorithms. One of our main products is a system called “SPOT”, which is currently helping clinicians worldwide to track the progress and treatment of chronic wounds. As part of this product, we reconstruct 3D models of the wound using the smartphone camera, creating a model that allows automated and accurate measurements and tracking the healing process. One of the many challenges in reconstructing a 3D model from a set of images taken at different angles is identifying and matching keypoints across multiple views. In this project, your goal is to train deep learning models to detect, describe and match keypoints and fine tune these models.

ICL
ICL, a global manufacturer of products based on unique minerals, fulfills humanity’s essential needs, primarily in three markets: agriculture, food and engineered materials. The experience, knowledge and professionalism derived from establishing Israel’s potash industry over a period of 80 years of intensive activity and major investment in R&D, have transformed ICL into a world leader in specialty fertilizers, bromine and flame retardants. ICL produces approximately a third of the world’s bromine, and is the world’s sixth largest potash producer, as well as one of its leading providers of pure phosphoric acid.

Project:
Predicting formation of dry areas in DSW evaporation ponds

ICL's Carnallite raw material production is carried out in the evaporation ponds. The ponds are used to precipitate Halite (NaCl) and carnallite (MgKCl). The available surface area of the evaporation ponds is a key factor in the efficiency and yield of the raw material precipitation. In order to maximize and optimize the ponds area, we must be aware of what areas are blocked, dry or nearly dry as it effects the raw material capacity and yield. In the last couple of years, we use LiDAR aerial bathymetric measurements to quantify the area loss in the ponds.The goal of this project is to predict the formation of dry areas in the Carnallite Ponds.

K-Health
Healthcare systems are collapsing around the world. Half the world population has limited or no access to decent healthcare. In the West, healthcare costs are rising, waiting times are unbearable, and trust in healthcare and even in the science of medicine is at an all time low. Brilliant doctors are overworked, burdened with liability and outdated protocols, and forced to try to understand complicated medical conditions in just a few minutes.

Project:
Diagnosis prediction

The K App allows people to insert their symptoms and see the medical conditions people like them had, based on clinical data and chat with a live physician. The doctor can diagnose them, prescribe medication and refer for further treatment. The data provided by K's physicians has tremendous potential to improve the diagnostic model. The goal of the project is to build a high performance classifier that predicts the diagnosis of the physician using the information collected in the visit. The model will be used for analysis and improvement of our current classification mechanism.

Medisafe
Medisafe is a personalized medication management company, developing technologies that help patients adhere to the prescribed treatment, and get better health outcomes. Our Just-In-Time-Interventions system provides patients with relevant information and interventions personalized to the patient journey. “Care Connector” system helps nurses to connect to patients and solve issues that require human touch.

Project:
Predicting patient engagement for successful and timely interventions

Medisafe, a personalized medication management platform, develops technologies that help patients to adhere to the prescribed treatment, and to get better health outcomes. Medisafe app is the main channel of communication with the patients so it’s crucial to make sure that it’s engaging and stays on device throughout the treatment. Churning from the app in most cases means churning from the treatment. The goal in this project is to create a churn prediction model which will detect users likely to disengage amd allow nurses to focus on the patients that are at higher risk, and to give them effective and timely interventions.

Nvidia
The demand for more computing power, efficiency and scalability is constantly accelerating in the HPC, Cloud, Web 2.0, Machine Learning, Data Analytics, and Storage markets. To address these demands Nvidia Networking business unit (Formerly Mellanox) provides complete end-to-end solutions supporting InfiniBand and Ethernet networking technologies. Nvidia NBU products include Smart Networking Devices, Switches, Interconnects and DPU’s. 

Project:
Prediction of chip failures at early stages of production flow

The purpose of this project is to build a model which predicts failures of manufactured chips based on indicators coming from the different stages of production. The data is coming from different components of the chip and from different stages of the manufacturing process. The goal of the project is to build a basic model which provides prediction of test failures based on early process and results and deals with hierarchical nature of the data

Roojoom
Roojoom is reinventing the way brands communicate with their customers. Our AI-based journey orchestration platform is driving individual customer towards the brand’s business goals, continuously throughout the whole customer lifecycle.  The world’s largest service providers such Comcast, Bouygues Telecom and others are among our early adopters’ customers.

Project:
Fault Resolution AI Agent

For customer service journeys, Roojoom's platform is used for driving troubleshoot & resolve (TS&R) journeys continuously from issue to resolution across sessions. Roojoom’s TS&R AI aims to provide an autonomous customer fault resolution journey capability, which optimizes resolution times and improves First Time-Right rates (solving the issue at the customer’s first attempt for getting assistance). Today, we use machine learning algorithms to predict which treatment will solve the customer’s problem. The goal of this project is to improve our resolution rate by incorporating the sequence of interactions with the customer so far into the prediction model.  

Similarweb
As the most trusted platform for measuring online behavior, millions of people rely on SimilarWeb insights on a daily basis as the ground truth for their knowledge of the digital world. We empower anyone — from the curious individual to the enterprise business leader — to make smarter decisions by understanding why things happen across the digital ecosystem. 

Project:
App Domain matching

In Similarweb we wish to provide comprehensive data on all digital assets of a company, and present it as a single entity; one of the most sought after features requested by customers is the ability to match one or more applications (from Google Play or Apple Store) to its respective web domains.  For example, matching facebook.com domain, the facebook app in google play store (or apple store), the actual company facebook, the relevant linkedin etc. The project will entail using already available data and developing an algorithm that will recommend a match between applications and domains.

Supersmart
Supersmart creates the leading validation and loss prevention solutions for scan&go and self-checkout services. Our products are used by retailers both in Israel and abroad.  Our product uses various data sources including weight, visual recognition, data extracted from the retailer’s point of sale system and user generated data such as information recorded on the user’s shopping app - all in order to help prevent revenue losses caused by shopper’s scanning errors or fraud activity​.

Project:
Detecting anomalous shopper behavior

The worlds shops are stuck in the past. We still need a cashier to go over our items at the checkout. In order to move forward, we need a solution that can ensure low to nonexistent loss rates, without human supervision. This project’s mission is to use data gathered during users’ shopping trips in order to create a graph of user behavior and detect suspicious activity which could be a result of fraud, honest mistake, or the existence of an anomalous item.

Taboola
Taboola is a content recommendation platform which helps users find relevant content online, matching them with news stories, articles, blogs, videos, apps, products and other content they’re likely to want to explore. The platform is built in a way that helps publishers drive engagement, advertisers to surface their content to the right audiences at scale, and people to discover content they may like but never knew existed.  

Project:
Cloaking Score for Taboola campaigns

Scammers have been known to use cloaking techniques to buy ad inventory on publisher sites, social media, and apps. Scammers that intend on circumventing review systems typically serve ads that fall into three main categories; misleading, inappropriate, or malicious. These scammers rely on cloaking to evade ad policies, which is hurting the relationship with strategic publishers and creating a bad experience for our users. By monitoring all Taboola advertisers’ campaigns constantly with a ML algorithm we will be able to solve challenges above giving Taboola automated tools to fight cloaking.

Tipalti
Tipalti’s end-to-end accounts payable platform provides the infrastructure for long-term success. Tipalti provides a best-in-class partner payments experience, delivering a white-labeled experience with fast payments, global coverage, rich payment method choice, and unparalleled partner visibility.

Project:
My RoboError translator

At Tipalti we are connecting to our customers ERPs in order to synchronize information from both of sides. When such a sync fails, we get an error message from the ERP side that tells us about the issue. For some messages we have a matching UI message which is displayed to the customer, explaininig what is the cause of the error, while for many others we don’t have a matching UI message - ending up with a generic error message. We’d like to solve this and improve user experience by creating an algorithm to classify the errors and produce a correct UI message for every ERP error.

Project:
Fradulent transaction detection

At Tipalti we are moving funds all over the world and in order to validate that activity we want to have an algorithm that monitors the daily transactions for anomalies and fradulent behavior. The mission in this project is to gather the historical information and create a way to spot patterns in the behavior in order to mark the anomalies accordingly, and to define fraud conditions (criteria) to meet in order to to give our customers a heads up whenever abnormal behavior takes place by creating anomaly and fraud likelihood metrics

TRG
We are a software development solution house focused on the Cyber Intelligence & Emergency Response domain,enableing our customers to offer better and secure lives to their citizens , by using our off-the-shelf integrated solutions that instantly make a real change.

Project:
Named Entity Recognition engine on Social Media

In order to deal with more sophisticated Advanced Persistent Threat (APT) attacks, it is indispensable to discover cybersecurity threat intelligence from unstructured or semi-structured data specifications. Named entity recognition tools allows us to extract and cluster in different levels specific topics and entities form public and social media datasets.  In this project you will work with anonymised social media posts and analyse them with NLP tools to perform sentiment analysis and topic extraction

TSG
TSG is a global provider of C4ISTAR, HLS and Cyber Security solutions with over 40 years of experience. TSG is privately held by Israel Aerospace Industries (IAI) and Formula Systems (50% each). From Strategic to Tactical level C4I solutions, implemented in various Armed Forces and Agencies. Air and Naval C4I, as well as HLS, Intelligence, Telecom and Cyber Security cutting edge solutions.

Project:
Adaptive XGBoost for deterioration prediction in ICU patients

We currently have in use an adaptive XGBoost algorithm that we use for transfer learning of patient deterioration models. It is used in Beilinson and Sheba hospitals to predict the deterioration of Covid-19 patients using transfer learning methods trained on clinical data from ICU patients (pre-Covid-19). These algorithms are a necessary part of the streaming of on-line medical prediction system, and in this project we are looking to improve our existing algorithm.

Project:
Early Prediction of Sepsis

Sepsis is a potentially life-threatening condition caused by the body's response to an infection. Sepsis occurs when the body's response to an infection is out of balance, triggering changes that can damage multiple organ systems. Septic shock (namely, infection throughout the body) is a potentially fatal medical condition that occurs when septic condition deteriorates further. The goals of this project is to predict sepsis onset before its happening and improve the predictive capabilities of our current model.

Weissbeerger
WeissBeerger is a product development tech company for the food, beverage, and retail industries. Made up of a team of industry experts, AI and machine learning engineers, and data scientists, today, after having been acquired by Anheuser-Busch InBev (BUD), Weissbeerger serves as an analytics and innovation arm of the largest brewery in the world. WeissBeerger has developed the leading data analytics solution for the Food & Beverage, and Retail Industries, enabling our grocery stores, pubs, and restaurants to optimize the way they run their businesses by leveraging the advantages of a data-driven world.

Project:
Dynamic Promotion Engine

We want to create an engine that finds the optimal promotion to offer at each time frame (e.g., week, month, quarter). This engine should rank the promotions by their marginal profit for each outlet and to the entire chain. The engine should return an evaluation matrix that can assess the promotion's expected and real performance. The results will be compared with previous promotions that we created using statistical and analytical methods.

Windward
Windward is a global leader in maritime and technological innovation, empowering actionable decisions through its Predictive Intelligence solutions. By fusing artificial intelligence, big data, and maritime expertise it enables clients and partners to understand the maritime ecosystem and its broader impact on security, compliance, safety, supply chain, and business.

Project:
Port mapping using behavioral vessel data

Ports are the departure and arrival points of almost all vessel journeys and play a major role in the supply chain of various goods. As such, ports and their associated areas are a central point of interest in the maritime world, and being able to identify them, locate them accurately and distinguish between them can be of great value. In this project we will use vessel behavioral data in order to map port related areas of interest.

Zebra Medical Vision
Zebra-Med’s mission is to provide radiologists the tools they need to make the next leap in patient care. The demand for medical imaging services is continuously increasing, outpacing the supply of qualified radiologists and stretching them to produce more output, without compromising patient care.

Project:
Pulmonary Embolism Identification

Acute Pulmonary Embolism (APE) is an urgent medical condition that should be diagnosed and treated ASAP. It is caused when a clot breaks free and enters the arteries in the lung. The goal of this project is to build a fast and accurate algorithm aimed at detection and classification of PE cases based on a Kaggle dataset of chest CTPA images. 

Timeline

  • 23.01.2021
    Deadline for submission of project priorities

  • 28.01.2021
    Publication of student assignment to projects

  • Feb 2021
    Data understanding, exploration, domain understanding

  • 05.03.2021
    Project presentation – Initial phase: Data, goals, tools

  • Mar-June 2021
    Main work on project – modelling, evaluation etc

  • July 2021
    Final project presentations

Contact Us

If you have any questions, you can find us in social networks or send a message to
y-data@yandex-team.com