Analytics India Magazine, in partnership with the Association of Data Scientists (ADaSci), has just concluded its two-day virtual conference Deep Learning DevCon 2020. Scheduled for 29th and 30th October, DLDC 2020 has brought together the pioneers and best minds of deep learning and ML industry from around the globe.
The deep learning conference of the year, DLDC is a leading virtual conference exclusively designed for deep learning practitioners across the world. The two-day conference has witnessed an exciting lineup of extraordinary speakers, interesting talks and practical hands-on workshops.
Here are the key highlights from the action-packed two-day deep learning conference:
Understanding Speech: Moving beyond ASRs
Abhinav Tushar, head of AI at the Bengaluru-based conversational AI startup Vernacular.ai spoke about the significance of speech and the hidden emotions behind it. He stated that speech is far different than text, and it is much more complicated. Tushar also shared some of the factors that are impacting the responses include content, environment, speaker characteristics and paralinguistics.
AI For Medical Imaging: What It Takes
Rohit Ghosh, the founder of Qure.ai, took this session at DLDC 2020, where spoke about the challenges of AI in healthcare and presented the audience with various aspects of medical imaging. Ghosh further explained that Qure.ai algorithms might require at least 3 million images to work well. Lack of open-source datasets is a hindrance to the algorithm developers.
Default Rate Prediction Models for Self- employment in Korea using Ridge, Random Forest and Deep Neural Network
This session was a paper presentation, given by Dongsuk Hong, Senior Researcher at Korea Credit Information Services(KCIS). In this session, she discussed ML and DL models for predicting self-employment default rates by utilising the credit information. Hong explained how they used micro-level variables such as loans, overdue history of individual businesses in Korean manufacturing sector with typical macro-economic ones in order to achieve performance enhancement in predicting default rates.
Musicians & Data scientists Create Hits with Ensembles
Next came an exciting session presented at the DLDC 2020 by Loveesh Bhatt and Arpita Sur, who is the senior manager of analytics and AVP / head cognitive computing system at Ugam respectively. In this session, the speakers discussed how data scientists could leverage stacked ensembles to improve the performance of predictive models.
Deep Learning For Tabular Data
Moving further ahead with DLDC 2020, Luca Massaron, the senior data scientist and Kaggle Master, spoke about what deep learning and deep neural networks are and why it is relevant. During his talk, Massaron shed light on the topic of deep learning for tabular data. He then discussed three challenges, which are data preparation, high cardinality and architecture. He explained several challenges that lie in tabular data such as mixed feature data-types, sparse data, etc.
Role of AI in steel industry / AI enabled manufacturing
DLDC 2020 also touched upon the understated field of manufacturing. The conference witnessed an interesting session with Ramesh Kumar, the head of analytics at Tata Steel. During the talk, Kumar shared how AI is being explored in the company and also spoke about some of the challenges that come into picture while deploying artificial intelligence. Kumar shared four significant challenges that intrude the AI function in the manufacturing sector — which are data, people, process and technology.
Anomaly Detection & Behaviour Analytics in Payments Fraud Risk Management
In another session, Satyam Misra, the head of Analytics and Insights group for Small Business Group at Intuit, spoke about how anomaly detection and behaviour analytics, in particular, is helping in the payments fraud risk management. Considering companies have been using traditional ways of risk management such as underwriting data, rule-based controls, credit checks, aggregation and more, Mishra spoke about how these methods come with flaws that make them inefficient for identifying and reacting to fraud patterns.
Who Priced my Insurance?
Continuing on the BFSI sector, DLDC 2020 also included a talk with Shilpi Bhabhra, head of the Analytics and Data Sciences at Acko General Insurance. In this session she spoke about how digital insurance was quickly changing the way risk is calculated and how the insurance amount ranged from known factors such as age and experience to colour of the car and the brand of the customer’s phone. The speaker’s talk was broadly divided into three main parts, which are background information on insurance, how ML is changing the game in the insurance sector and disruptive new-age technology.
Decade of ML Progress
Kaggle Grandmaster Konstantin Yakovlev spoke at length about how the ML community is lacking the creativity of late. In his talk, Konstantin underlined the importance of resource management rather than just focusing on moving the goalposts. “Today, most of ML is focussing on creating frameworks and not on resource management,” he added.
InnovFaceNet: Deep Face Recognition for Industrial Environments
In another paper presentation, DLDC 2020 called for two speakers — Nagarjun Gururak, Data Evangelist and Kanika Batra, Head of Data Science at Innovators Bay. In this session, they spoke about their research paper where they proposed a facial recognition technology that works real-time in any industrial setup to eliminate manual supervision, ensuring authorised access to the personnel in the plant. The proposed technology of facial recognition claimed to identify people that are wearing mask without training or fine-tuning the original model with real-time processing of 20 FPS on a CPU and an F1-score of 95.07%.
Stock Price Prediction Using Deep Learning Models
This paper presented a suite of deep learning-based models for stock price prediction. The speaker, Jaydip Sen, Professor at Praxis Business School used the historical records of the NIFTY 50 index listed in the National Stock Exchange (NSE) of India for building and testing the models. The proposition includes two regression models built on convolutional neural networks (CNNs), and three long-and-short-term memory (LSTM) network-based predictive models.
Is it in production?
In a session with Github, Pulkit Agarwal and Vinod Joshi discussed the various challenges of setting up an ML pipeline. Pulkit, who is part of the product team at Github, began by defining what MLOps is really about and what makes it challenging while organisations have figured out how to work with DevOps. The session further included four key challenges one might face while setting up an ML pipeline, which are collaboration on code, remote training, model bookkeeping and managing data code and updates.
Applied AI for IoT- enabled mobility platform
The speaker Sanup Haridas, who heads the analytics and data science team in Vogo, gave an insight into how Vogo’s bike-rental platform operates. He mentioned that with the application of AI, IoT, and Bluetooth technology has made this platform more intelligent, reduced costs and improved user experience. This session, in particular, chartered Vogo journey from being dependent on manual supervision to moving to automate several aspects such as estimating the fuel consumption and avoiding several fraudulent practices.
Graph-based Embeddings to Optimize Website Segmentation for Digital Ad Campaigns
Continuing the paper presentations, this session was hosted by Dushyant Rai Tara, Senior Data Scientist at MiQ. In this talk, he presented his research paper where he talked about a data-driven approach to create a new contextual strategy from web traffic data that segments websites into groups for targeting. He then discussed his process by comparing with different techniques using a heuristic criterion to identify the most optimal method for vector representation.
Imbalance Handling with Combination of Deep Variational Autoencoder and NEATER
Divye Singh, who is a deep learning researcher, talked about the class imbalance problem which encounters different areas ranging from medical diagnosis to anomaly detection. In this paper presentation, Divye mentioned that this imbalanced class distribution makes extracting useful information very challenging for many popular algorithms. In this situation, optimising the overall accuracy can positively skew the predictions toward the majority class label.
Production of 3D Bioartificial “Organoids” Using Quantum-Enabled Deep Learning Neural Networks
Talking about organoids, This session was presented by Raul Villamarin Rodriguez, who is the Dean of the School of Business at Woxsen University. The session discussed the current situation of organ failure and how quantum enabled deep learning neural networks can help in mitigating the issue. As per the solution proposed by the speaker, carried out in collaboration with his team, this problem can be solved by producing 3D bioartificial organoids. These organoids are created using quantum-enabled CNN techniques. The speaker spoke of the various algorithms used such as — evolutionary, genetic, and quantum algorithms which not only speed up the process but also ensure greater accuracy.
Credit Risk Scoring for Unbanked: Opportunities and Challenges with AI/ML
Mounika Mydukur, head of analytics at mPokket took the audience through some of the challenges they face with credit scoring and how they leverage deep AI and ML technologies to overcome them and identify new opportunities, at DLDC 2020. In her talk, she highlighted the challenges from mPokket’s perspective — an instant loan app designed to provide loans to college students and young working professionals. Mydukur further shared some of the alternate data points to assess credit risk for the unbanked.
Covid-19 TeleHealth Solution
Presented by Bhagvan Kommadi, Director at ValueMomentum, this session discussed deep learning and how it is evolving in various areas like knowledge management, gene regulation, genome organisation, and mutation effects. Kommadi explained how deep learning is helping in identifying disease symptoms, undiagnosable disease analysis and detecting introgression. Further, he talked about how the technology can also assist in estimating historical recombination rates, identifying selective sweeps, and estimating demography of population genetics.
S&P 500 Stock’s Movement Prediction using Deep Learning
Last but not least, DLDC 2020 organised a paper presentation by the Senior Manager at Quinnox Rahul Gupta. In this session, he presented his paper, where he talked about predicting the movement of stock consisting of the S&P 500 index. Gupta discussed the various methods to predict the stock movement and being used in the market currently for algorithm trading and alpha-generating systems using traditional mathematical approaches.