Mediaboss Marketing

Center director Murali Annavaram (second from right) and associate director Meisam Razaviyayn (far right) hosted the event (PHOTO CREDIT: USC Viterbi)

On May 23rd, the USC-Meta Center for Research and Education in AI and Machine Learning hosted its first major event. The center was established in Fall of 2021 with the goal of addressing the technological challenges related to making AI and machine learning sustainable, efficient, and scalable. Nearly 100 attendees including, Meta representatives and USC Viterbi students and faculty who work in the AI and Machine Learning spaces, attended the workshop and poster session.

Our center identified four pillars of support we can provide to the next generation of AI and Machine Learning researchers and industry leaders, said Murali Annavaram, Professor of Electrical and Computer Engineering and the centers inaugural director. Those pillars, research, teaching, fellowships, and outreach, were all addressed at this inaugural in-person event.

Research

Seventeen USC Viterbi students shared their research in AI and Machine Learning with visitors from Meta (PHOTO CREDIT: USC Viterbi)

The research focus of the event centered on a series of ML presentations on the efficiency, security, and privacy of machine learning algorithms, followed by an in-depth poster session. A total of 25 PhD students shared their research with the visiting Meta team. These events allowed Meta visitors to get a better understanding of the breadth and depth of work already being done at USC Viterbi in AI and Machine Learning. The face-to-face nature of the session allowed industry representatives to ask questions and collaborate with researchers on future ideas.

USC Viterbi already has a strong presence in AI and Machine Learning research, which is why we were selected to establish this center, said Meisam Razaviyayn, Assistant Professor of Industrial and Systems Engineering, and the centers Associate Director. The research sessions are a perfect opportunity to showcase our strength to our Meta partners and to establish future collaborations.

The event was a great opportunity to connect with folks in industry and exchange ideas. It was nice to present my work to people who might be able to apply our algorithms in practice to help protect peoples privacy, at Meta and beyond, said PhD student and presenter Andrew Lowy.

Teaching

As the field of AI and Machine Learning grows and technology continues to improve, teaching and courses that address this area can and must evolve alongside it. At this workshop, the USC-Meta center announced two important curriculum enhancements that will better prepare students for research and work in these fields. The electrical and computer engineering department is adding systems and implementation-oriented ML course offerings in the near future (pending university approvals) and the industrial and systems engineering department is significantly enhancing the curriculum of the Masters in Analytics program.

Fellowships

Center leaders announced the inaugural cohort of six new MS fellowship recipients, a first of its kind at USC. In fact, these fellowships represent the first comprehensive, all-tuition paid fellowships for MS students with diverse and rigorous academic backgrounds. The awards also will foster diversity and inclusion among MS students working in relevant areas of research.

The generous support of the USC Meta Center has provided an invaluable opportunity for Masters students to pursue their passion for AI, said Camillia Lee, USC Viterbi Associate Dean of Graduate Admission. The USC Meta fellowship will play a major role in achieving the Viterbi Schools goal of continuing to attract some of the most talented and diverse students in the country, and prepare them for future careers in AI and machine learning.

Outreach

During the centers board meeting, USC Viterbi and Meta discussed the importance of improving mentorship and support to students who are part of the center or supported by center fellowships. The team agreed that outreach in this area should be expanded beyond PhD students to also include MS, undergraduate students, and K-12 students.

In collaboration with USC Viterbis SURE (summer undergraduate research experience) program, the center will support multiple undergraduate student researchers visiting USC this summer. That program, which sees students from all over the country come to campus to be exposed to engineering research, will now enjoy more support from the USC-Meta center. Five high school students will also be supported and mentored by the center this summer as part of USC Viterbis SHINE program. Leaders of both those programs were quick to praise the centers support.

This type of support, mentorship, and encouragement is transformative for students who are underrepresented in STEM, said Katie Mills, co-director of the USC Viterbi K-12 STEM Center. SHINE alumni go on to leading research universities, majoring in STEM, with a strong sense of the benefits to society of research and technology. The USC Meta Center is making those opportunities more available at the local level.

The additional financial support for SURE from the USC-Meta center will allow us to bring even more students into this program and let us focus even more on the important and growing fields of AI and Machine learning, said Andy Jones-Liang, associate director of academic services.

Despite the already great progress made in the nascent 5 months since the inception of the center, the center directors have a vision to bring academic and industry researchers in ML to collaborate on compelling societal challenges and to provide mentoring support for the next generation of ML engineers.

Published on June 16th, 2022

Last updated on June 16th, 2022

Follow this link:
USC-Meta Center Brings Progress to AI Education and Research - USC Viterbi | School of Engineering - USC Viterbi School of Engineering

Datatonic, a leader for Data + AI consulting on Google Cloud, announced it has received the 2021 Google Cloud Specialization Partner of the Year award for Machine Learning.

Datatonic was recognized for the companys achievements in the Google Cloud ecosystem, helping joint customers scale their Machine Learning (ML) capabilities with Machine Learning Operations (MLOps) and achieve business impact with transformational ML solutions.

Datatonic has continuously invested in expanding their MLOps expertise, from defining what good MLOps looks like, to helping clients make their ML workloads faster, scalable, and more efficient. In just the past year, they have built high-performing MLOps platforms for global clients across the Telecommunications, Media, and e-Commerce sectors, enabling them to seamlessly leverage MLOps best practices across their teams.

Marketing Technology News:Pega Named a Leader in Real-Time Interaction Management by Independent Research Firm

Their recently open-sourced MLOps Turbo Templates, co-developed with Google Clouds Vertex AI Pipelines product team, showcase Datatonics experience implementing MLOps solutions, and Google Clouds technical excellence to help teams get started with MLOps even faster.

Were delighted with this recognition from our partners at Google Cloud. Its amazing to see our team go from strength to strength at the forefront of cutting-edge technology with Google Cloud and MLOps. Were proud to be driving continuous improvements to the tech stack in partnership with Google Cloud, and to drive impact and scalability with our customers, from increasing ROI in data and AI spending to unlocking new revenue streams. Louis Decuypere CEO, Datatonic

Marketing Technology News:MarTech Interview with Venkat Nagaswamy, CMO at Mitel

Google Cloud Specializations recognize partner excellence and proven customer success in a particular product area or industry, said Nina Harding, Global Chief, Partner Programs and Strategy, Google Cloud. Based on their certified, repeatable customer success and strong technical capabilities, were proud to recognize Datatonic as Specialization Partner of the Year for Machine Learning.

Datatonic is a data consultancy enabling companies to make better business decisions with the power of Modern Data Stack and MLOps. Its services empower clients to deepen their understanding of consumers, increase competitive advantages, and unlock operational efficiencies by building cloud-native data foundations and accelerating high-impact analytics and machine learning use cases.

Originally posted here:
Datatonic Wins Google Cloud Specialization Partner of the Year Award for Machine Learning - MarTech Series

Founded in 2008 by entrepreneur brothers Chintan and Harit Soni at IIM Ahmedabad's Centre for Innovation, Incubation and Entrepreneurship in India, Ecolibrium provides expert advisory as well as technology-driven sustainability solutions to enable businesses in commercial and industrial real estate to reduce energy consumption and ultimately achieve their net zero carbon ambitions.

Relocating its global headquarters to London, Ecolibrium has raised $5m in a pre-Series A funding round as it looks to expand its international footprint to the UK. The round was co-led by Amit Bhatia's Swordfish Investments and Shravin Bharti Mittal's Unbound venture capital firm, alongside several strategic investors.

Ecolibrium launches in the UK today having already signed its first commercial contract with Integral, JLL's UK engineering and facilities service business.

The fundraising and UK expansion builds on Ecolibrium's considerable success in Asia Pacific, where its technology is being used across 50 million sq ft by more than 150 companies including Amazon, Fiat, Honeywell, Thomson Reuters, Tata Power, and the Delhi Metro. An annual reduction of 5-15% in carbon footprint has been achieved to date by companies which have deployed Ecolibrium's technology.

Ecolibrium has also strengthened its senior UK management team, as it prepares to roll-out its green platform across the UK, by hiring facilities and asset management veteran Yash Kapila as its new head of commercial real estate. Kapila previously held senior leadership positions with JLL across APAC and EMEA regions.

Introducing SmartSense

At the heart of Ecolibrium's offer is its sustainability-led technology product SmartSense, which assimilates thousands of internet of things (IoT) data points from across a facility's entire energy infrastructure.

This information is then channelled through Ecolibrium's proprietary machine learning algorithms, which have been developed over 10 years by their in-house subject matter experts. Customers can visualise the data through a bespoke user interface that provides actionable insights and a blueprint for achieving operational excellence, sustainability targets, and healthy buildings.

This connected infrastructure generates a granular view of an asset's carbon footprint, unlocking inefficiencies and empowering smart decision-making, while driving a programme of continuous improvement to deliver empirical and tangible sustainability and productivity gains.

Preparing for future regulation

Quality environmental data and proof points are also providing a distinct business advantage at this time of increasing regulatory requirements that require corporates to disclose ESG and sustainability performance. Ecolibrium will work closely with customers to lead the way in shaping their ESG governance.

According to Deloitte, with a minimum Grade B Energy Performance Certification (EPC) requirement anticipated by 2030, 80% of London office stock will need to be upgraded an equivalent of 15 million sq ft per annum.

Research from the World Economic Forumhas found that the built environment is responsible for 40% of global energy consumption and 33% of greenhouse gas emissions, with one-fifth of the world's largest 2,000 companies adopting net zero strategies by 2050 or earlier. Technology holds the key to meeting this challenge, with Ecolibrium and other sustainability-focused changemakers leading the decarbonisation drive.

Chintan Soni, Chief Executive Officer at Ecolibrium, said:"Our mission is to create a balance between people, planet and profit and our technology addresses each of these objectives, leading businesses to sustainable prosperity. There is no doubt the world is facing a climate emergency, and we must act now to decarbonise and protect our planet for future generations.

"By using our proprietary machine learning-led technology and deep in-house expertise, Ecolibrium can help commercial and industrial real estate owners to deliver against ESG objectives, as companies awaken to the fact that urgent action must be taken to reduce emissions and achieve net zero carbon targets in the built environment.

"Our goal is to partner with companies and coach them to work smarter, make critical decisions more quickly and consume less. And, by doing this at scale, Ecolibrium will make a significant impact on the carbon footprint of commercial and industrial assets, globally."

The UK expansion has been supported by the Department for International Trade's Global Entrepreneur Programme. The programme has provided invaluable assistance in setting up Ecolibrium's London headquarters and scaling in the UK market.

In turn, Ecolibrium is supporting the growth of UK innovation, promoting green job creation, and providing tangible economic benefits, as part of the country's wider transition to a more sustainable future.

Minister for Investment Lord Grimstone said: "Tackling climate change is crucial in our quest for a cleaner and green future, something investment will play an important part in.

"That's why I'm pleased to see Ecolibrium's expansion to the UK. Not only will the investment provide a revolutionary sustainability solution to reduce carbon emissions across various sectors, it is a continued sign of the UK as a leading inward investment destination, with innovation and expertise in our arsenal".

About Ecolibrium

Ecolibrium is a machine learning-led decarbonisation platform balancing people, planet and profit to deliver sustainable prosperity for businesses.

Founded in 2008 by entrepreneur brothers Chintan and Harit Soni, Ecolibrium provides expert advisory as well as technology-driven sustainability solutions to enable commercial and industrial real estate owners to reduce energy consumption and ultimately achieve their net zero carbon ambitions.

Ecolibrium's flagship technology product SmartSense is currently being used across 50 million sq ft by more than 150 companies including JLL, Amazon, Fiat, Honeywell, Thomson Reuters, Tata Power, and the Delhi Metro. SmartSense collects real-time information on assets, operational data and critical metrics using internet of things (IoT) technology. This intelligence is then channelled through Ecolibrium's proprietary machine learning algorithms to visualise data and provide actionable insights to help companies make transformative changes to their sustainability goals.

For more information, visit: http://www.ecolibrium.io

For press enquiries, contact: FTI Consulting: [emailprotected], +44 (0) 2037271000

Photo -https://mma.prnewswire.com/media/1837227/Ecolibrium_Yash_Kapila_and_Chintan_Soni.jpg

SOURCE Ecolibrium

See the original post:
Machine learning-led decarbonisation platform Ecolibrium launches in the UK - PR Newswire

Raman, A. P., Anoma, M. A., Zhu, L., Rephaeli, E. & Fan, S. Passive radiative cooling below ambient air temperature under direct sunlight. Nature 515, 540544 (2014).

ADS CAS PubMed Article Google Scholar

Zhai, Y. et al. Scalable-manufactured randomized glass-polymer hybrid metamaterial for daytime radiative cooling. Science 355, 10621066 (2017).

ADS CAS PubMed Article Google Scholar

Li, P. et al. Large-scale nanophotonic solar selective absorbers for high-efficiency solar thermal energy conversion. Adv. Mater. 27, 45854591 (2015).

CAS PubMed Article Google Scholar

Kumar, R. & Rosen, M. A. Thermal performance of integrated collector storage solar water heater with corrugated absorber surface. Appl. Therm. Eng. 30, 17641768 (2010).

Article Google Scholar

Planck, M. The Theory of Heat Radiation (P. Blakinstons Son & Co., 1914).

MATH Google Scholar

Zhu, J., Hsu, C. M., Yu, Z., Fan, S. & Cui, Y. Nanodome solar cells with efficient light management and self-cleaning. Nano Lett. 10, 19791984 (2010).

ADS CAS PubMed Article Google Scholar

Zhou, L., Yu, X. & Zhu, J. Metal-core/semiconductor-shell nanocones for broadband solar absorption enhancement. Nano Lett. 14, 10931098 (2014).

ADS CAS PubMed Article Google Scholar

Lee, B. J., Chen, Y. B., Han, S., Chiu, F. C. & Lee, H. J. Wavelength-selective solar thermal absorber with two-dimensional nickel gratings. J. Heat Transfer 136, 17 (2014).

Google Scholar

Yin, X., Yang, R., Tan, G. & Fan, S. Terrestrial radiative cooling: Using the cold universe as a renewable and sustainable energy source. Science 370, 786791 (2020).

ADS CAS PubMed Article Google Scholar

Nie, X. et al. Cool white polymer coatings based on glass bubbles for buildings. Sci. Rep. 10, 110 (2020).

ADS Article CAS Google Scholar

Mandal, J. et al. Hierarchically porous polymer coatings for highly efficient passive daytime radiative cooling. Science 362, 315319 (2018).

ADS CAS PubMed Article Google Scholar

Zhang, H. et al. Biologically inspired flexible photonic films for efficient passive radiative cooling. Proc. Natl. Acad. Sci. 117, 202001802 (2020).

Google Scholar

Krishna, A. et al. Ultraviolet to mid-infrared emissivity control by mechanically reconfigurable graphene. Nano Lett. 19, 50865092 (2019).

ADS CAS PubMed Article Google Scholar

Sala-Casanovas, M., Krishna, A., Yu, Z. & Lee, J. Bio-inspired stretchable selective emitters based on corrugated nickel for personal thermal management. Nanoscale Microscale Thermophys. Eng. 23, 173187 (2019).

ADS CAS Article Google Scholar

Sullivan, J., Yu, Z. & Lee, J. Optical analysis and optimization of micropyramid texture for thermal radiation control. Nanoscale Microscale Thermophys. Eng. https://doi.org/10.1080/15567265.2021.1958960 (2021).

Article Google Scholar

Campbell, P. & Green, M. A. Light trapping properties of pyramidally textured surfaces. J. Appl. Phys. 62, 243249 (1987).

ADS Article Google Scholar

Leon, J. J. D., Hiszpanski, A. M., Bond, T. C. & Kuntz, J. D. Design rules for tailoring antireflection properties of hierarchical optical structures. Adv. Opt. Mater. 5, 18 (2017).

Google Scholar

Zhang, T. et al. Black silicon with self-cleaning surface prepared by wetting processes. Nanoscale Res. Lett. 8, 15 (2013).

ADS CAS Article Google Scholar

Liu, Y. et al. Hierarchical robust textured structures for large scale self-cleaning black silicon solar cells. Nano Energy 3, 127133 (2014).

CAS Article Google Scholar

Dimitrov, D. Z. & Du, C. H. Crystalline silicon solar cells with micro/nano texture. Appl. Surf. Sci. 266, 14 (2013).

ADS CAS Article Google Scholar

Peter Amalathas, A. & Alkaisi, M. M. Efficient light trapping nanopyramid structures for solar cells patterned using UV nanoimprint lithography. Mater. Sci. Semicond. Process. 57, 5458 (2017).

Article CAS Google Scholar

Mavrokefalos, A., Han, S. E., Yerci, S., Branham, M. S. & Chen, G. Efficient light trapping in inverted nanopyramid thin crystalline silicon membranes for solar cell applications. Nano Lett. 12, 27922796 (2012).

ADS CAS PubMed Article Google Scholar

Rahman, T., Navarro-Ca, M. & Fobelets, K. High density micro-pyramids with silicon nanowire array for photovoltaic applications. Nanotechnology 25, 485202 (2014).

PubMed Article CAS Google Scholar

Singh, P. et al. Fabrication of vertical silicon nanowire arrays on three-dimensional micro-pyramid-based silicon substrate. J. Mater. Sci. 50, 66316641 (2015).

ADS CAS Article Google Scholar

Zhu, J. et al. Optical absorption enhancement in amorphous silicon nanowire and nanocone arrays. Nano Lett. 9, 279282 (2009).

ADS PubMed Article CAS Google Scholar

Wei, W. R. et al. Above-11%-efficiency organic-inorganic hybrid solar cells with omnidirectional harvesting characteristics by employing hierarchical photon-trapping structures. Nano Lett. 13, 36583663 (2013).

ADS CAS PubMed Article Google Scholar

Peng, Y. J., Huang, H. X. & Xie, H. Rapid fabrication of antireflective pyramid structure on polystyrene film used as protective layer of solar cell. Sol. Energy Mater. Sol. Cells 171, 98105 (2017).

CAS Article Google Scholar

Sai, H., Yugami, H., Kanamori, Y. & Hane, K. Solar selective absorbers based on two-dimensional W surface gratings with submicron periods for high-temperature photothermal conversion. Sol. Energy Mater. Sol. Cells 79, 3549 (2003).

CAS Article Google Scholar

Deinega, A., Valuev, I., Potapkin, B. & Lozovik, Y. Minimizing light reflection from dielectric textured surfaces. J. Opt. Soc. Am. A 28, 770 (2011).

ADS Article Google Scholar

Shore, K. A. Numerical methods in photonics, by Andrei V. Lavrinenko, Jesper Laegsgaard, Niles Gregersen, Frank Schmidt, and Thomas Sondergaard. Contemporary Physics vol. 57 (2016).

Malkiel, I. et al. Plasmonic nanostructure design and characterization via deep learning. Light Sci. Appl. 7, 18 (2018).

CAS Article Google Scholar

Bojarski, M. et al. End to End Learning for Self-Driving Cars. 19 (2016).

Hinton, G. et al. Deep neural networks for acoustic modeling in speech recognition: The shared views of four research groups. IEEE Signal Process. Mag. 29, 1617 (2012).

Article Google Scholar

Spantideas, S. T., Giannopoulos, A. E., Kapsalis, N. C. & Capsalis, C. N. A deep learning method for modeling the magnetic signature of spacecraft equipment using multiple magnetic dipoles. IEEE Magn. Lett. 12, 15 (2021).

Article Google Scholar

Xiong, Y., Guo, L., Tian, D., Zhang, Y. & Liu, C. Intelligent optimization strategy based on statistical machine learning for spacecraft thermal design. IEEE Access 8, 204268204282 (2020).

Article Google Scholar

Zhang, C. A Statistical Machine Learning Based Modeling and Exploration Framework for Run-Time Cross-Stack Energy Optimization (University of North Carolina at Charlotte, 2013).

Book Google Scholar

Zhu, W. et al. Optimization of the thermophysical properties of the thermal barrier coating materials based on GA-SVR machine learning method: Illustrated with ZrO2doped DyTaO4system. Mater. Res. Express 8, 125503 (2021).

ADS CAS Article Google Scholar

Zhang, T. et al. Machine learning and evolutionary algorithm studies of graphene metamaterials for optimized plasmon-induced transparency. Opt. Express 28, 18899 (2020).

ADS CAS PubMed Article Google Scholar

Li, X., Shu, J., Gu, W. & Gao, L. Deep neural network for plasmonic sensor modeling. Opt. Mater. Express 9, 3857 (2019).

ADS CAS Article Google Scholar

Baxter, J. et al. Plasmonic colours predicted by deep learning. Sci. Rep. 9, 119 (2019).

ADS Article CAS Google Scholar

He, J., He, C., Zheng, C., Wang, Q. & Ye, J. Plasmonic nanoparticle simulations and inverse design using machine learning. Nanoscale 11, 1744417459 (2019).

CAS PubMed Article Google Scholar

Sajedian, I., Kim, J. & Rho, J. Finding the optical properties of plasmonic structures by image processing using a combination of convolutional neural networks and recurrent neural networks. Microsyst. Nanoeng. 5, 18 (2019).

Article Google Scholar

Han, S., Shin, J. H., Jung, P. H., Lee, H. & Lee, B. J. Broadband solar thermal absorber based on optical metamaterials for high-temperature applications. Adv. Opt. Mater. 4, 12651273 (2016).

CAS Article Google Scholar

Seo, J. et al. Design of a broadband solar thermal absorber using a deep neural network and experimental demonstration of its performance. Sci. Rep. 9, 19 (2019).

ADS Article CAS Google Scholar

Nadell, C. C., Huang, B., Malof, J. M. & Padilla, W. J. Deep learning for accelerated all-dielectric metasurface design. Opt. Express 27, 27523 (2019).

ADS CAS PubMed Article Google Scholar

Deppe, T. & Munday, J. Nighttime photovoltaic cells: Electrical power generation by optically coupling with deep space. ACS Photon. https://doi.org/10.1021/acsphotonics.9b00679 (2019).

Article Google Scholar

Ma, W., Cheng, F. & Liu, Y. Deep-learning-enabled on-demand design of chiral metamaterials. ACS Nano 12, 63266334 (2018).

CAS PubMed Article Google Scholar

Li, Y. et al. Self-learning perfect optical chirality via a deep neural network. Phys. Rev. Lett. 123, 16 (2019).

CAS Google Scholar

Balin, I., Garmider, V., Long, Y. & Abdulhalim, I. Training artificial neural network for optimization of nanostructured VO2-based smart window performance. Opt. Express 27, A1030 (2019).

ADS CAS PubMed Article Google Scholar

Elzouka, M., Yang, C., Albert, A., Prasher, R. S. & Lubner, S. D. Interpretable forward and inverse design of particle spectral emissivity using common machine-learning models. Cell Rep. Phys. Sci. 1, 100259 (2020).

Article Google Scholar

Peurifoy, J. et al. Nanophotonic particle simulation and inverse design using artificial neural networks. arXiv 18 (2017). https://doi.org/10.1117/12.2289195.

An, S. et al. A Deep learning approach for objective-driven all-dielectric metasurface design. ACS Photon. 6, 31963207 (2019).

See the rest here:
Deep learning based analysis of microstructured materials for thermal radiation control | Scientific Reports - Nature.com

Youre at the wheel of your car but youre exhausted. Your shoulders start to sag, your neck begins to droop, your eyelids slide down. As your head pitches forward, you swerve off the road and speed through a field, crashing into a tree.

But what if your cars monitoring system recognised the tell-tale signs of drowsiness and prompted you to pull off the road and park instead? The European Commission has legislated that from this year, new vehicles be fitted with systems to catch distracted and sleepy drivers to help avert accidents. Now a number of startups are training artificial intelligence systems to recognise the giveaways in our facial expressions and body language.

These companies are taking a novel approach for the field of AI. Instead of filming thousands of real-life drivers falling asleep and feeding that information into a deep-learning model to learn the signs of drowsiness, theyre creating millions of fake human avatars to re-enact the sleepy signals.

Big data defines the field of AI for a reason. To train deep learning algorithms accurately, the models need to have a multitude of data points. That creates problems for a task such as recognising a person falling asleep at the wheel, which would be difficult and time-consuming to film happening in thousands of cars. Instead, companies have begun building virtual datasets.

Synthesis AI and Datagen are two companies using full-body 3D scans, including detailed face scans, and motion data captured by sensors placed all over the body, to gather raw data from real people. This data is fed through algorithms that tweak various dimensions many times over to create millions of 3D representations of humans, resembling characters in a video game, engaging in different behaviours across a variety of simulations.

In the case of someone falling asleep at the wheel, they might film a human performer falling asleep and combine it with motion capture, 3D animations and other techniques used to create video games and animated movies, to build the desired simulation. You can map [the target behaviour] across thousands of different body types, different angles, different lighting, and add variability into the movement as well, says Yashar Behzadi, CEO of Synthesis AI.

Using synthetic data cuts out a lot of the messiness of the more traditional way to train deep learning algorithms. Typically, companies would have to amass a vast collection of real-life footage and low-paid workers would painstakingly label each of the clips. These would be fed into the model, which would learn how to recognise the behaviours.

The big sell for the synthetic data approach is that its quicker and cheaper by a wide margin. But these companies also claim it can help tackle the bias that creates a huge headache for AI developers. Its well documented that some AI facial recognition software is poor at recognising and correctly identifying particular demographic groups. This tends to be because these groups are underrepresented in the training data, meaning the software is more likely to misidentify these people.

Niharika Jain, a software engineer and expert in gender and racial bias in generative machine learning, highlights the notorious example of Nikon Coolpixs blink detection feature, which, because the training data included a majority of white faces, disproportionately judged Asian faces to be blinking. A good driver-monitoring system must avoid misidentifying members of a certain demographic as asleep more often than others, she says.

The typical response to this problem is to gather more data from the underrepresented groups in real-life settings. But companies such as Datagen say this is no longer necessary. The company can simply create more faces from the underrepresented groups, meaning theyll make up a bigger proportion of the final dataset. Real 3D face scan data from thousands of people is whipped up into millions of AI composites. Theres no bias baked into the data; you have full control of the age, gender and ethnicity of the people that youre generating, says Gil Elbaz, co-founder of Datagen. The creepy faces that emerge dont look like real people, but the company claims that theyre similar enough to teach AI systems how to respond to real people in similar scenarios.

There is, however, some debate over whether synthetic data can really eliminate bias. Bernease Herman, a data scientist at the University of Washington eScience Institute, says that although synthetic data can improve the robustness of facial recognition models on underrepresented groups, she does not believe that synthetic data alone can close the gap between the performance on those groups and others. Although the companies sometimes publish academic papers showcasing how their algorithms work, the algorithms themselves are proprietary, so researchers cannot independently evaluate them.

In areas such as virtual reality, as well as robotics, where 3D mapping is important, synthetic data companies argue it could actually be preferable to train AI on simulations, especially as 3D modelling, visual effects and gaming technologies improve. Its only a matter of time until you can create these virtual worlds and train your systems completely in a simulation, says Behzadi.

This kind of thinking is gaining ground in the autonomous vehicle industry, where synthetic data is becoming instrumental in teaching self-driving vehicles AI how to navigate the road. The traditional approach filming hours of driving footage and feeding this into a deep learning model was enough to get cars relatively good at navigating roads. But the issue vexing the industry is how to get cars to reliably handle what are known as edge cases events that are rare enough that they dont appear much in millions of hours of training data. For example, a child or dog running into the road, complicated roadworks or even some traffic cones placed in an unexpected position, which was enough to stump a driverless Waymo vehicle in Arizona in 2021.

With synthetic data, companies can create endless variations of scenarios in virtual worlds that rarely happen in the real world. Instead of waiting millions more miles to accumulate more examples, they can artificially generate as many examples as they need of the edge case for training and testing, says Phil Koopman, associate professor in electrical and computer engineering at Carnegie Mellon University.

AV companies such as Waymo, Cruise and Wayve are increasingly relying on real-life data combined with simulated driving in virtual worlds. Waymo has created a simulated world using AI and sensor data collected from its self-driving vehicles, complete with artificial raindrops and solar glare. It uses this to train vehicles on normal driving situations, as well as the trickier edge cases. In 2021, Waymo told the Verge that it had simulated 15bn miles of driving, versus a mere 20m miles of real driving.

An added benefit to testing autonomous vehicles out in virtual worlds first is minimising the chance of very real accidents. A large reason self-driving is at the forefront of a lot of the synthetic data stuff is fault tolerance, says Herman. A self-driving car making a mistake 1% of the time, or even 0.01% of the time, is probably too much.

In 2017, Volvos self-driving technology, which had been taught how to respond to large North American animals such as deer, was baffled when encountering kangaroos for the first time in Australia. If a simulator doesnt know about kangaroos, no amount of simulation will create one until it is seen in testing and designers figure out how to add it, says Koopman. For Aaron Roth, professor of computer and cognitive science at the University of Pennsylvania, the challenge will be to create synthetic data that is indistinguishable from real data. He thinks it is plausible that were at that point for face data, as computers can now generate photorealistic images of faces. But for a lot of other things, which may or may not include kangaroos I dont think that were there yet.

Excerpt from:
Is fake data the real deal when training algorithms? - The Guardian