Scientific program

Dec 07-08, 2022    Chicago, USA

4th International Conference on

Big Data, AI and IoT

  • Home -
  • Scientific program

Keynote Forum

Mark DeSantis

Mark DeSantis

Bloomfield Robotics, USA

Title: Deep Learning and Crop Inspection: Bigger Yields, Better Harvests and Safer Crops

Abstract:

Since the dawn of agriculture, crop monitoring and inspection remain a mainstay of every farmer’s routine. Today, many farmers visually inspect their crops armed with a variety of tools to help ensure ideal plant health and performance. Although human visual inspection remains an essential part of agriculture, it has many challenges and many limitations. Research over the last decade or so has assessed the applicability of computer vision and deep learning to address the crop inspection challenge [Nusk2011, Nusk2014, Blom2009, Herr2015]. These approaches have shown tremendous promise, they are only just now beginning to go beyond the research phase into commercialization. Around 2015, image processing methods using deep neural networks began to replace the earlier classical computer vision approach, providing both better performance and more generalizable results. Again, through the early work of the CMU team, a StalkNet [Bawe2018] architecture was developed, which combines an RCNN feature detector with a GAN-based pixel segmenter. To date, StalkNet has been trained to measure dozens of widely varying features in different crops, ranging from leaf necrosis to fruit ripeness to sorghum seed size for grain yield. The first market Bloomfield has chosen to address is grape growing and vineyard inspection, but we see CEA as a natural next step in the progression of our technology and a large opportunity. Flash combines high-resolution flash-lighted stereo RGB images with a cloud-based deep learning pipeline to inspect and assess the health and performance of every plant in a field or grow, one plant at a time. The result, so far, with Bloomfield’s vineyard customers, is yield estimation, pest/disease detection, labor-saving, and digitalization. This comprehensive analysis forms the foundation for Bloomfield’s health and performance assessment of each geo-located plant, one plant at a time through a web-based dashboard accessible via tablet, cellphone, or computer. Bloomfield’s approach to inspecting and assessing plants contrasts sharply with the visual inspection which includes sparse subjective judgments of randomly sampled plant data.

Biography:

Mark DeSantis is a serial tech entrepreneur, lecturer, and educator always looking for interesting things to do. 

Title: Keynote Speaker Slots Available

Speakers

Karishma Nanda

Karishma Nanda

Bournemouth University, UK

Title: Neural Network Based Prediction in Recommender

Abstract:

This paper aims to contribute to the cold start problem in recommender system with Neural Network based approach. There are several attempts in academia and in the industry to improve the recommender system. For instance, latent matrix factorization is an algorithm that solves the recommendation problem, it produces efficient outcomes from the core problem. Latent factors are not directly observed but are inferred from other factors. It can be computed by assuming a specific number of such factors and then transforming the large user-item matrix into a smaller matrix based on previously assumed factors. These smaller matrices can be multiplied to reproduce a close approximation to the original user-item matrix using a technique called matrix factorization. Assuming that the matrix can be written as the product of two low-rank matrices, matrix factorization techniques seek to retrieve missing or corrupted entries. Matrix factorization approximates the matrix entries by a simple fixed-function — namely, the inner product — acting on the corresponding row and column latent feature vectors. Substituting a neural architecture for the inner product that learns from the data, improves recommendation problem and deals with the cold start problem

Biography:

Karishma Nanda has her expertise in AI and passion for Neural networks. Her open and contextual evaluation creates new pathways for improving Recommender System. System-Architect with 5 years of experience involving E2E design, development, and implementation of applications built on Pega. Currently pursuing Masters in Artificial Intelligence & Data Science. I have Strong database, Programming skills, machine learning algorithms, data visualization concepts. Possess good communication and interpersonal skills, self-motivated, a quick learner, a team player, have a good understanding of Agile methodology. Specialty in designing Smart Systems based on MAPE-K and utilizing AI and Data Science. Looking for a challenging role where I can utilize my knowledge of system design, AI, Data Science, Machine Learning.

Sergio Mastrogiovanni

Sergio Mastrogiovanni

Nubiral, USA

Title: Intelligent Healthcare

Abstract:

Statement of the Problem: In 2020, COVID-19 exposed the fragility of the health sector. In the US in particular, the most expensive healthcare system in the world, it also faces a tremendous challenge in responding to diagnostic needs. One of the biggest challenges in medical imaging like MRI is not the high cost per se, but the capacity. An MRI session lasts between 15 and 60 minutes. There are hospitals with only one device or even no one. Medical imaging is one of the best use cases for AI in healthcare, but lack of physician engagement and data bottlenecks can make the technology less useful than promised. When used to decode the complicated nature of MRIs, CT scans, and other testing modalities, advanced analytical tools have proven their ability to extract meaningful information to improve decision-making, sometimes with greater precision than humans themselves. With deep learning, it is possible to capture fewer data and thus scan faster, while preserving or even enhancing the rich information content of MRI images. The key is to train artificial neural networks to recognize the underlying structure of the images to fill in the missing views from the accelerated scan. This approach is similar to how humans process sensory information. When we experience the world, our brains often receive an incomplete image, as in the case of darkened or dimly lit objects, that we need to convert into actionable information.

Biography:

Sergio Mastrogiovanni is a senior data scientist, executive, entrepreneur, AI evangelist, and data storyteller with career success leveraging advanced data analytics and technology integration to boost sustainable revenue, inspire high-performing teams and manage change through digital transformation and continuous improvement, and his passion in this world is about making data accessible to people. He teaches Intelligent Automation at NYU and is the Head of Data and Innovation at Nubiral. Strong expertise in developing simulation, optimization, cost reduction, and risk assessment models and deploying business analytics and process automation solutions. Masters in Analytics, NYU Stern MBA, Certified Six Sigma Black Belt, certified MIT AI practitioner, certified RPA developer, Microsoft Certified System Engineer, Azure Certified Data Scientist, Data Engineer, and AI Associate. AWS Certified Big Data professional, Innovation coach, Columbia Data Scientist, and visualization Zen that won awards on innovation, leadership, and process improvement. Fluent in Spanish, English, and Portuguese.

Title: Speaker Slots Available

Keynote Forum

Masoud Atael

Masoud Atael

York University, Canada

Title: A Hybrid Convolutionary Neural Network and Low-rank Tensor Learning Algorithm for Tensor-on-Tensor Regression

Abstract:

The problem of predicting a set of tensorial outputs based on inputs of tensor form has been receiving increasing attention in recent years. This problem arises in various areas of mathematical, statistical, and computational sciences, and generalizes the case of the widely used scalar-on-scalar regression methods. In this paper, we develop a tensor-on-tensor regression framework using a hybrid of coevolutionary neural networks and low-rank tensor learning algorithms. Our proposed framework integrates several promising approaches which have been developed previously to tackle this problem and extends their domain of applications. In particular, we demonstrate the advantage of this framework in comparison with traditional methods through an example of predicting the third-order tensors which arise within the procedures required for performing the time-homogeneous top-K ranking algorithm. Computational results are further provided which pertain to the analysis of the U.S. stock market during the time period from January 1990 to December 2019.

Biography:

Masoud Atael is a Solution Specialist at SAS and completed masters at York University, Canada.

Title: Keynote Speaker Slots Available

Speakers

Darius Burschka

Darius Burschka

University of Munich, Germany

Title: Discussion on Explainable AI for Robotic Applications

Abstract:

The current AI approaches based on Deep Learning were originally developed for fast data queries in large datasets for search engines, social media, and advertising. The common property of these fields is that they are not used in critical decision loops (control) of a robotic system, but they serve as an index key to finding previously searched information that is similar to the current situation. This origin resulted in a strong development of the data labeling direction that is essential for fast data association. In my talk, I want to discuss the necessary extensions that need to be added to the current AI approaches to make them applicable for decisions on robotic systems. While the approaches become increasingly better in answering the "what is there?" question, a robotic system requires in addition also information about the "confidence" of each query. A 95% accurate system running for 24 hours fails during 72min/day. The control system needs to identify these periods to prevent damages to the system and the surrounding environment. Additionally, usually, not a single sensor is used for control, and for a robust data-fusion, a (metric) error covariance is important. I show ways how to achieve this goal in the DL context. The last step is a discussion of temporal extensions of the current AI approaches, which need to understand not only the current snapshot of the scene but its temporal evolution to grasp the current context and model dynamic events. I will present our initial work on temporal scene modelling and discuss the necessary updates to the benchmarking in current AI to make it applicable to robotics

Biography:

Darius Burschka received his Ph.D. degree in Electrical and Computer Engineering in 1998 from the Technische Universitätt München in the field of vision-based navigation and map generation with binocular stereo systems. In 1999, he was a Postdoctoral Associate at Yale University, Connecticut, where he worked on laser-based map generation and landmark selection from video images for vision-based navigation systems. From 1999 to 2003, he was an Associate Research Scientist at the Johns Hopkins University, Baltimore, Maryland.  Later 2003 to 2005,  he was an Assistant Research Professor in Computer Science at the Johns Hopkins University. Currently, he is a Professor in Computer Science at the Technische Universität München, Germany, where he heads the Machine Vision and Perception group, he is a member of the Scientific Board of the Munich School for Robotics and Machine Intelligence (MSRM).  His areas of research are sensor systems for mobile and medical robots and human-computer interfaces. The focus of his research is on vision-based navigation and three-dimensional reconstruction from sensor data.  He is a Senior Member of IEEE.

Selma Elizabeth Blum

Title: Artificial Intelligence: Technology Applied on Criminal Justice

Title: Speaker Slots Available