NVIDIA AI Conference includes a mixture of keynotes, speakers, and instructor-led labs across a range of topics. Browse the at-a-glance agenda below. Check back often for updates as we confirm more speakers.
NVIDIA AI Conference includes a mixture of keynotes, speakers, and instructor-led labs across a range of topics. Browse the at-a-glance agenda below. Check back often for updates as we confirm more speakers.
Deep Learning for Scene Description Generation
Deep Learning for Healthcare (Medical Imaging)
Learn how to leverage deep neural networks (DNN) within the deep learning workflow to solve a real-world image classification problem using NVIDIA DIGITS. You will walk through the process of data preparation, model definition, model training and troubleshooting. You will use validation data to test and try different strategies for improving model performance using GPUs. On completion of this lab, you will be able to use DIGITS to train a DNN on your own image classification application.
Level: Beginner
Data, Tools and Frameworks used: Jupyter Notebook, DIGITS, Caffe, MNIST, CNN
Prerequisite: Familiarity with DL concepts will be helpful but this is not required.
This lab explores various approaches to the problem of semantic image segmentation, which is a generalization of image classification where class predictions are made at the pixel level. We use the Sunnybrook Cardiac Data to train a neural network to learn to locate the left ventricle on MRI images. In this lab, you will learn how to use popular image classification neural networks for semantic segmentation, how to extend Caffe with custom Python layers, become familiar with the concept of transfer learning and train two Fully Convolutional Networks (FCNs).
Level: Beginner
Data, Tools and Frameworks used: MRI, Sunnybrook Cardiac Dataset, DIGITS, Caffe, Python, FCN, CNN
Prerequisite: Familiarity with DIGITS, Python and DL concepts will be helpful but this is not required.
Convolutional neural networks (CNNs) have proven to be just as effective in visual recognition tasks involving non-visible image types as regular RGB camera imagery. One important application of these capabilities is medical image analysis, where we wish to detect features indicative of medical conditions and use them to infer patient status. In addition to processing non-visible imagery, such as CT scans and MRI, these applications often require us to process higher dimensionality imagery that may be volumetric and have a temporal component. In this lab you will use the deep learning framework MXNet to train a CNN to infer the volume of the left ventricle of the human heart from a time-series of volumetric MRI data. You will learn how to extend the canonical 2D CNN to be applied to this more complex data and how to directly predict the ventricle volume rather than generating an image classification. In addition to the standard Python API, you will also see how to use MXNet through R, which is an important data science platform in the medical research community.
Level: Intermediate
Data, Tools and Frameworks used: MRI, Sunnybrook Cardiac Dataset, NDSB2, MXnet, R, CNN
Prerequisite: Familiarity with R and DL concepts is strongly advised. Familiarity with MXnet will be helpful but this is not required.
The primary purpose here is to explore how deep learning can be leveraged in a healthcare setting to predict severity of illness in patients based on information provided in electronic health records (EHR). In this lab we will use the python library pandas to manage dataset provided in HDF5 format and deep learning framework Keras to build recurrent neural networks (RNN). In particular, this lab will construct a special kind of deep recurrent neural network that is called a long-short term memory network (LSTM). The general idea here is to develop an analytic framework powered by deep learning techniques that provides medical professionals the capability to generate patient mortality predictions at any time of interest. Such a solution provides essential feedback to clinicians when trying to assess the impact of treatment decisions or raise early warning signs to flag at risk patients in a busy hospital care setting. Finally, we will compare the performance of this LSTM approach to standard mortality indices such as PIM2 and PRISM3 as well as contrast alternative solution formulations using more traditional machine learning methods like logistic regression.
Level: Intermediate
Data, Tools and Frameworks used: PIM2, PRISM3, EHR, HDF5, CHLA Dataset, Keras, Theano, Python, RNN, LSTM
Prerequisite: Familiarity with Keras, Dataframes and DL concepts is strongly advised.
Effective descriptions of content within images and video clips has been performed with convolutional and recurrent neural networks. Users will apply a deep learning technique via a framework to create captions on data and generate their own captions.
Convolutional and recurrent neural networks can be combined to generate effective descriptions of content within images and video clips. In this TensorFlow lab, attendees will learn about data processing and preparation for network ingestion, network configuration, training, and inference. At the end the lab, participants will know how to ingest and process input data from images as well as sentences, extract image feature vectors from a pretrained network, one-hot encode sentences, concatenate input data, configure and RNN and train it, and then perform inference with their own trained network.
Level: Intermediate
Data, Tools and Frameworks used: Jupyter Notebook, Python, TensorFlow, MSCOCO, CNN, RNN
Prerequisite: Familiarity with TensorFlow, Python and DL concepts will be helpful but this is not required.
Effective descriptions of content within images and video clips has been performed with convolutional and recurrent neural networks. Users will apply a deep learning technique via a framework to create captions on data and generate their own captions.
Convolutional and recurrent neural networks can be combined to generate effective descriptions of content within images and video clips. In this TensorFlow lab, attendees will learn about data processing and preparation for network ingestion, network configuration, training, and inference. At the end the lab, participants will know how to ingest and process input data from images as well as sentences, extract image feature vectors from a pretrained network, one-hot encode sentences, concatenate input data, configure and RNN and train it, and then perform inference with their own trained network.
Level: Intermediate
Data, Tools and Frameworks used: Jupyter Notebook, Python, TensorFlow, MSCOCO, CNN, RNN
Prerequisite: Familiarity with TensorFlow, Python and DL concepts will be helpful but this is not required.
AI is transforming cities. Municipalities around the world are realizing the benefits to public safety and city operations that can be acheived through the power of Deep Learning and GPUs. NVIDIA's Metropolis platform makes it easy for customers to develop and deploy AI-based solutions to smart and safe city challenges, and to extract valuable insights from the hundreds of millions of cameras deployed around the world. <br><br>In this talk we will show how AI is being applied to video analytics and why it achieves better results than traditional solutions. We will talk about the resources that NVIDIA provides to accelerate the deployment of these solutions. And we will discuss NVIDIA's AI City partner program and the benefits that it provides to the industry.
Consumer web sites we all use every day have applied deep learning & AI for several years to recognize and organize massive amounts of video and audio data. In the enterprise, however, lack of supported deep learning software and applications have often been a barrier to adoption. Today, with all the major deep learning frameworks supported by NVIDIA’s DGX family of deep learning supercomputers and by the NVIDIA GPU Cloud, and with major enterprise applications like SAP rolling out new deep-learning enabled features, there is no reason to wait to start your first deep learning project.
Consumer web sites we all use every day have applied deep learning & AI for several years to recognize and organize massive amounts of video and audio data. In the enterprise, however, lack of supported deep learning software and applications have often been a barrier to adoption. Today, with all the major deep learning frameworks supported by NVIDIA’s DGX family of deep learning supercomputers and by the NVIDIA GPU Cloud, and with major enterprise applications like SAP rolling out new deep-learning enabled features, there is no reason to wait to start your first deep learning project.
The AI Revolution is here, and goes beyond automation; huge opportunity exists for both productivity & consumption gains
AI (Artificial Intelligence) is very hot now, why it is so? What are the AI activities we are doing and what are the promising applications with AI? This talk will introduce our AI activities especially deep learning related development for smart sensing in various applications surveillance, autonomous, logistics, factory automation, retails, etc,
Deep Learning has raised Artificial Intelligence to another level because of Big data and super powerful computation, with better robustness and higher accuracy compared to traditional machine learning. Deep Learning Statement: Accuracy is improved from 95% to 99% with the good robustness, means a “Game Changing”, to make impossible be possible, to derive super intelligence beyond human brain! By providing some application example it will help to understand the power of Artificial Intelligence.
This talk will also introduce Deep Reinforcement Learning with application examples in self-driving car to show one of the future trends for Deep Learning development, how to move from supervised learning to unsupervised or self-supervised learning.
This presentation summarizes the current state of Artificial Intelligence adoption by Enterprise; discusses the barriers to adoption; suggests practical approaches; and demonstrates rapidly implemented and mature use cases that immediately realize ROI.
Augmented Intelligence focuses on democratizing Artificial Intelligence in a symbiotic approach to enable non-technical staff to immediately harness data science capabilities. The approach immediately result in increased productivity.
As a leading visual search and recognition technology provider, ViSenze has indexed more than several hundred million of images and support millions of image search query every day. We'll discuss various aspects of how we conduct deep learning based product design, model training and GPU optimisation for online and offline processing. We will also talk about how we use HydraVision, Visenze's internal deep learning training platform, to manage the end to end training workflow and improve the R&D iteration.
Deep Instinct is the first company to provide Deep Learning neural network detection and prevention capability for zero day cyber and unknown threats. Deep Instinct will provide a detailed example of the major performance delta between Machine Learning and Deep Learning cyber detection capability.
Robotics science and technology have evolved from the seminal applications in industrial robotics for manufacturing to today’s varied applications in service, health care, education, entertainment and other industries including construction, mining and agriculture. One common theme in these emerging applications is the human-centered nature in unstrucrtured environments, where robotic systems surround humans, aiding and working with us to enrich and enhance the quality of our lives. This talk reviews the state-of-the-art developments in fundamental capabilities in both electro-mechanical hardware and artificial intelligence. An example on a self-driving cars is presented. This talk will then conclude with the challenges in science and technology to further accelerate the robotics revolution.
A framework that combines the use of data analytics, machine learning and optimization can solve real-world complex problems effectively. In particular, the use of spatial-temporal data to build predictive models can help derive optimal plans. In this presentation, Prof Lau will shed light on the application of this framework, illustrated through two real-world problems: (1) in public safety, where crime data can be used to predict crime occurrence and in turn, be used for police officer staffing; and (2) in maritime traffic safety, where AIS (Automatic Identification System) data can be used to predict vessel movement trajectories, to provide timely advice for vessels navigating through the Singapore Strait. The latter is an ongoing projecting with Maritime and Port Authority of Singapore, in collaboration with Fujitsu and A*STAR.
Glaucoma is a blinding ocular disorder with no cure that is extremely difficult to detect in the earliest stages. Glaucoma is responsible for 40% cases of blindness in Singapore and has a prevalence that reaches 12% among those aged over 70 years. Standard glaucoma diagnosis is complex, expensive and requires time and a battery of multiple clinical tests: a robust, fast and automated diagnosis tool for early glaucoma detection has potential enormous societal impacts. Optical Coherence Tomography (OCT) scans are inexpensive, only require a few minutes to obtain, and can provide detailed and high-resolution 3D images of the optic nerve head, the main site of glaucoma damage. In this talk, I will describe how the Deep-Learning methodology can be leveraged for analyzing 3D OCTs and providing robust probabilistic glaucoma diagnostics.
Serving 120 million Indonesian consumers who are constantly changing their behavior is not an easy task.
At Lippo Group, one of the largest business conglomerates in Indonesia, we’re focused on impacting lives through the development of well-planned, sustainable business execution, and quintessential customer experience
In order to better understand our customers, we’ve developed a comprehensive digital ecosystem. In this talk, we’ll discuss how we are able to integrate data from multiple lines of business across several industries into one analytic platform, and how our mantra of “deep and fast analytics” is opening up new opportunities for improved customer engagement. We’ll describe how we were able to connect thousands of data sources and gain immediate deep analytic insight from mining this massive amount of data. Through rigorous data modeling, we are able to gain a much better understanding of our customers, leading to increased customer engagement and satisfaction.
Information is one of the most important asset in any digitalized industry. In order to protect sensitive data, and to meet regulatory requirements imposed by different jurisdictions, documents in any regulated institutions must be monitored, classified, categorized and labelled internally. In order to facilitate this in the age of 'big data', machine learning and algorithmic based solutions are the future of data management and governance.
Machine learning has recently achieved very promising results in a wide range of areas such as computer vision, speech recognition and natural language processing. It aims to learn hierarchical representations of data using deep learning models. Videos captured from cameras are automatically processed and analysed. Deep learning algorithms applied to video analytics in used cases for object detection, face recognition , image classification and image captioning.
This presentation will introduce the research at the ROSE Lab in general and our use of deep learning, as well as more traditional approaches, for intelligent video analytics.
Automatically searching for optimal hyperparameter configurations is of crucial importance for applying deep learning algorithms in practice. Recently, Bayesian optimization has been proposed for optimizing hyperparameters of various machine learning algorithms. However, probabilistic surrogates require accurate estimates of sufficient statistics (e.g., covariance) of the error distribution and thus need many function evaluations with a sizeable number of hyperparameters. This makes them inefficient for optimizing hyperparameters of deep learning algorithms, which are highly expensive to evaluate. In this talk, I will present a novel deterministic and efficient hyperparameter optimization method that employs radial basis functions as error surrogates. The proposed mixed integer optimization algorithm, called HORD, searches the surrogate for the most promising hyperparameter values through dynamic coordinate search and requires many fewer function evaluations. Extensive evaluations on MNIST and CIFAR-10 for four deep neural network types demonstrate HORD significantly outperforms the well-established Bayesian optimization methods such as GP, SMAC, and TPE. I will conclude the talk with a practical demonstration of hyperparameter optimization using HORD.
Condition-Based Maintenance (CBM) is a form of Artificial Intelligence, which is also known as Predictive Maintenance (PM), Predictive Asset Maintenance (PAM), or Industrial Internet-of-Things (IIoT). CBM scenarios usually evolve around expensive and important machinery. A successful CBM process for a machine can help with increasing asset life, preventing failures, aid in planning for resources and materials, and reduce maintenance cost and production downtime. In order to benefit from CBM, a constant monitoring and recording of the machine status data is required. In this talk, an introduction on CBM is provided, followed by CBM case studies in escalators and trains.
Automatic generation of caption to describe the content of an image has been gaining a lot of research interests recently, where most of the existing works treat the image caption as pure sequential data. Natural language, however possess a temporal hierarchy structure, with complex dependencies be- tween each subsequence. In this paper, we propose a phrase- based hierarchical Long Short-Term Memory (phi-LSTM) model to generate image description. In contrast to the conventional solutions that generate caption in a pure sequential manner, our proposed model decodes image caption from phrase to sentence. It consists of a phrase decoder at the bottom hierarchy to decode noun phrases of variable length, and an abbreviated sentence decoder at the upper hierarchy to decode an abbreviated form of the image description. A complete image caption is formed by combining the generated phrases with sentence during the inference stage. Empirically, our proposed model shows a better or competitive result on the Flickr8k, Flickr30k and MS-COCO datasets in comparison to the state-of-the art models. We also show that our proposed model is able to generate more novel captions (not seen in the training data) which are richer in word contents in all these three datasets.
Machine learning (ML) research often operates within silos, separate from the people who created the data and disconnected from its original purpose. Published datasets add value to the ML community, yet ML research outcomes are rarely incorporated by the data generators themselves.
As one of Australia’s largest data generators, Monash University seeks to cultivate ML communities that are embedded in the data generation process, from experiment design to analysis. By upskilling researchers with intimate knowledge of their data in applied ML techniques and enabling access to technology, we believe that an ML-centric mindset will unlock unseen potential in data insights and operational efficiencies. This presentation will showcase our efforts to raise awareness of GPU accelerated ML approaches powered by the MASSIVE supercomputer, and novel modes of access via our “Strudel” and “Strudel Web” HPC desktops application. We show two projects that have seen benefits from ML approaches to data analysis through proactive outreach: ASpirin in Reducing Events in the Elderly (ASPREE), a joint study attracting over US$50m in funding between the US and Australia, and the largest prevention trial ever conducted in Australia; and NHMRC-funded research in X-ray video analysis of rabbit kitten breathing as an analogue to human infant respiratory development.
We will show preliminary outcomes demonstrating how these initiatives build lasting partnerships with researchers, opening dialogue and encouraging novel collaborative ML approaches to research data analysis.
During this talk, we'll review statistical learning approaches for predicting molecular phenotypes.
We'll review strategies in building these models, and how to apply them in disease association studies and clinical trials. We'll discuss machine and Bayesian learning approaches, and how to leverage evidence from existing studies and knowledge-bases. We'll motivate and demonstrate these approaches in predicting nicotine metabolism, an important biomarker of smoking cessation and lung cancer.
Reinforcement Learning (RL) techniques have attracted a lot of attention recently due to its role in achieving human expert level skills in games such as Go and a number of video games. However, there are also important applications of RL techniques in engineering domains, such as autonomous vehicles, robotics, manufacturing and communication and computing systems, which require fast and safe responses, and are usually large scale systems. This presentation covers the application of RL techniques in engineering settings and touches on how recent developments in deep learning and GPU acceleration have further enabled solutions that cannot be obtained through other means.
The success of Alibaba's e-commerce along the past decade has manifested the power of data intelligence. Alibaba’s underlying data intelligence technology, cloud computing capabilities, artificial intelligence algorithms, and combined with customized hardware, such as GPU, can be applied to numerous sectors beyond ecommerce, from urban development to manufacturing, even to environment and agriculture. We will touch upon typical scenarios in e-commerce sector and elaborate recent progress in ET City Brain, ET Industry Brain and ET Environment Brain, in particular the advancement of real-time video analytics. Finally we will close off with future outlook on the rollout of data intelligence.
With rapid rise and increase of Big Data and AI as a new breed of high-performance workloads on supercomputers, we need to accommodate them at scale, and thus the need for R&D for HW and SW Infrastructures where traditional simulation-based HPC and BD/AI would converge. The TSUBAME3 supercomputer at Tokyo Institute of Technology became online in Aug. 2017, and became the greenest supercomputer in the world on the Green 500 at 14.11 GFlops/W; the other aspect of TSUBAME3, is to embody various BYTES-oriented features to allow for HPC to BD/AI convergence at scale, including significant scalable horizontal bandwidth as well as support for deep memory hierarchy and capacity, along with high flops in low precision arithmetic for deep learning. TSUBAM3's technologies will be commoditized to construct one of the world’s largest BD/AI focused open and public computing infrastructure called ABCI (AI-Based Bridging Infrastructure), hosted by AIST-AIRC (AI Research Center), the largest public funded AI research center in Japan. The performance of the machine is slated to be well above 130 Petaflops for machine learning, but again the true nature of the machine is being BYTES-oriented, with acceleration in I/O and other data-centric properties desirable for accelerating BD/AI. ABCI will be online 1H 2018 and will serve as a reference architecture for datacenters for such convergence.
Over the years, Dell EMC has been working on democratizing HPC technology to help business and industries of all size to leverage new technologies as part of their digital transformation. Today, the emerging technologies around AI, powered by HPC, allow industries to even transform deeper. Join us to learn more about what Dell EMC AI strategy to help researchers as well as industries to benefit from those new technologies.
This presentation will discuss some of the key challenges facing airports today and how Changi Airport adopted a SMART airport approach that leverages on sensors, data fusion and cognition capabilities to enable solutions in Changi Airport. Through this session, we hope to share the necessary paradigm shift towards a more dynamic approach to problem solving in today’s constrained operating environment.
CSIRO, Australia’s national science agency, we innovate for tomorrow while delivering impact today – for our customers, all Australians and the world. CSIRO Scientific Computing provides the computing services that underpin the delivery of every project the organisation undertakes. This talk will look at where AI and Deep Learning fit into the High Performance Computing CSIRO’s service offering that enables the science impact of Australia’s largest science organisation. How does CSIRO manage the newer services around AI/Deep Learning alongside the more traditional CPU and queue based delivery methods? What are teams that are needed to support these platforms and the science projects that have directly benefited from having these teams in place to assist their science.
Data is the fuel driving rapid innovation powered by artificial intelligence. Enterprises need a modern data hub purpose-built for machine learning, accelerating insight while simplifying complex data pipelines in analytics. And the data hub needs to be massively parallel at its core and has the power to deliver unprecedented performance and simplicity for data scientists and engineers. From small, metadata-heavy workloads to random, large file accesses, your organization needs to prepare for the unknown. Learn how FlashBlade and DGX are engineered to deliver maximum performance in any machine learning environment, reducing data processing time from days to hours.
An introduction to IBM's PowerAI, a purpose-built bundling of systems and software components for hardware-accelerated deep learning analytics.
Gartner reports that, by 2020, approximately 50% of enterprises will be actively using AIOps (Algorithmic IT Operations) techniques to provide insight into both business execution and IT Operations, which is an increase from fewer than 10% today. We will discuss how machine learning techniques can help to predict occurrence of the service outages of data center in advance so that the IT operation team can act proactively to prevent the outages.
Protect Sponsorship Business Value by Measuring What You Pay For via automated video analytics powered by machine learning.
There are many metropolitan cities around the world and most of these cities are facing challenges from multiple fronts such as public safety, operational efficiency, rising population, traffic problems and so on. AI-based computer vision allows us to observe and visualize our cities in a clearer, more concise and holistic way, to design and build our cities for the greatest efficiency and react to situations in a more timely manner. In this talk we will introduce existing and new technologies in AI-based computer vision and its applications in smart cities, for example, Aerial Imagery super-resolution and defogging for better visibility, Structured information for city design and operation, Crowd behavior and traffic management.
This presentation will discuss how you deal with human interfaces at scale on the classification, training and deployment of AI at the edge.
Global trends of urbanization, digitization, and industrialization are the drive towards smarter cities. The smart city’s computational infrastructure requires the ability to process megatrend datasets and visualize rich features within a given energy budget. GPU-accelerated data centers provide the hardware and software platform to achieve the smart cities objective. We review practical steps to developing smart city applications including object detection, facial recognition, and anomaly detection. Finally, these practices are tied to broader issues of privacy and transparency and possible solutions in the algorithmic space.
Data scientists, developers, and senior decision makers are facing a rapidly evolving landscape for their AI applications. They must harness an exponentially increasing amount of data and pick the right algorithms to deliver competitive advantage to their organization. However, the challenges don’t stop there. Infrastructure matters when you’re deploying deep learning/AI or accelerated analytics solutions. In this session you will learn how the Cisco GPU-Accelerated Data Center can help you design the best platform for you needs.
In this talk, we will introduce the FoodAI project, a pioneering AI project in Singapore for promoting smart food consumption and healthy lifestyle towards smart nation. Specifically, the goal of the FoodAI project is to build state-of-the-art food image recognition technologies for smart food logging using deep learning and GPUs. We will introduce the research and development of FoodAI technologies, present case studies with real users, and highlight some open research topics.