Data Transformers Podcast

Episode Title: Can Video Artificial Intelligence help with use cases like Elderly care and Child care?

Episode Summary:

Video AI is evolving and evolving rapidly in many segments such as healthcare for diagnostic purposes, MarTech for analyzing videos for brand recognition and Ad placement for example. Video AI usage in elderly care and child care are ripe for huge benefits as they require significant human participation. Video AI can address both costs as well as skill shortage in those areas. Personalization and analyzing consumer behavior are segments evolving for video AI usage. Still barriers exist in compute performance for modelling 3D world, real-time computing and inferencing, and barriers in power consumption especially in edge AI where there is limited power. While dealing with AI ethics, it is very important to separate privacy related issues from bias related issues. Bias is inherently in human beings and not in technology. So technology should be used to de-bias decisions.

00:43: Motivation behind jumping from one role to another for example moving to Tivo

02:31: Progression of role of data over the years. The scale and sophistication has increased. 

04:32: Focus on customer value delivered in innovation. Amazon is the extreme example of delivering customer value

08:30: Role of video in healthcare specifically in diagnostic applications; Excellent in tightly constrained applications

11:20: Computer vision in martech is evolving. Examples detecting brands in video or objects in videos for ad placement

12:15: Personalization and customer sentiment analysis are evolving subjects

14:12: Barriers to rapid advancements. Ex: computing performance for 3D world models.

15:02: Real time computing and inference is also a barrier. Power consumption on the edge is also an issue to be resolved.

17:39: Ethics in AI and privacy. 2 very hot topics. Customers have gotten used to security cameras but not for behavioral analysis

19:00: Bias is a totally different issue. Technology by itself is not biased but people implementing technology are biased.

21:00: Elderly care and child care are very human intensive tasks. But technology like Video AI can be very helpful in helping manage that.

Resources mentioned in this episode:

AV8.Ai Community 

Call to action: If you are interested in Video AI, Join AV8.AI

Data Transformers Podcast

Join Peggy and Ramesh as they explore the exciting world of Data Management, Data Analytics, Data Governance, Data Privacy, Data Security, Artificial Intelligence, Cloud Computing, Internet Of Things.

Is Video Artificial Intelligence Ready for Prime Time?

‘

Episode Title: Is Video AI Ready For Prime Time?

Episode Summary:

Video AI is a growing market with lots of innovation. The Video AI market encompasses Video surveillance, Automatic/self-drive vehicles, content moderation in video, automatic video editing. Convolution Neural Networks is the backbone of Video AI in many applications and the challenge lies in training the data as well as abstracting the outcomes for better outpost. The field is still emerging and the technology is still evolving in many of these areas. As an example, even though Youtube would like to have a general understanding of the video so they know when to insert relevant ads, the technology to do that is just emerging. Dale Hitt, who worked at many startups as well as large companies, goes over his experience with innovation at small and large companies.

(02:29) What is ADAS? Video intelligence in cars.

(03:10) Computer vision in sports

(04:43): Cambria technology – How it helps Vision AI

(05:58): Video intelligence in at-home fitness equipment

(08:11) Training data for video applications. Challenges with limited data and unrelated data

(10:20): Limitations of convolution neural networks and need for abstraction layers instead of more variables

(12:13): Various market segments for video AI applications

(14:37): Security surveillance market segment

(15:18): Video understanding in Youtube for monetization

(17:21): AV8.AI, a community for Video AI innovators for all market segments like content moderation, automatic video editing, self-driving vehicles

(19:29): Missile guidance systems to AI startups – Professional journey

(21:52): Startup journeys – From VC backed to bootstrapped – what is it like?

(23:19): Lessons from Tivo startup journey

(24:53): Challenges of innovating at large companies

(25:59): Transferable lessons from startups

Resources mentioned in this episode:

https://AV8.Ai Community

Data Transformers Podcast

Join Peggy and Ramesh as they explore the exciting world of Data Management, Data Analytics, Data Governance, Data Privacy, Data Security, Artificial Intelligence, Cloud Computing, Internet Of Things.

How to address the blind spots of data scientists

‘

Episode Title : How to address the blind spots of data scientists

Episode Summary:

A good data scientist is not only good at coding, tweaking models but is also good at assessing the outputs of the models in the context of business decisions. As data scientists spend upto 80% of their time data munging, they will be better off spending some quality upfront time with the business leaders asking questions about the customer journey and how the data is collected along the way. Even though the data scientists are not expected to be business domain experts, they should think of their output in the context of business outcomes and communicate their results in business terms. The episode also emphasizes the need for all types and sizes of companies to get acquainted with analytics or fall behind.

B2C company Versus B2B company (01:00): Having come from a B2C company like Disney, Phil had to adapt to dealing with challenges of dealing with B2B company like Teradata. Each day could be different at Teradata as Phil would be talking one day to a video games company, next day a healthcare company, and a transportation company the next day. So Phil needed to learn to switch hats quickly.

Business focus first, Data focus next (04:00): Majority of a data scientist’s time (upto 80%) is spent on data wrangling and a maximum of 20% on actual model development & tweaking. So a data scientist’s time is well spent initially talking to the business units on 3 things: (1) What data is collected (2) What is the customer journey (3) When and how is the data collected along the journey? The more time spent in clarifying these questions, the better it is for the team.

Blind spots of data scientists (09:02): Majority of the data scientists do not focus on the business processes or attempt to understand the business domain and that is a mistake. Yes, it is true that the businesses know their domains better than anyone else but the data scientist should attempt to put the data in the context of the business domain and not look at it independently.

A solution that is appropriate for their business (11:06): Businesses should evaluate technology solutions which are relevant for their business based on their current state of business. Phil gives an example of hiring an expert from the casino industry while at NBC Universal and the solution proposed was ahead of its time and not suitable for NBC. And that was a mistake.

Analytics best practices (15:12): Every company should look at using analytics in their business or otherwise they’ll be left out. They can start with a small test data set but should start somewhere. Engage with an external company on a small project to get upto speed before building a data team. Then the business can start assessing if they want to continue relying on external teams or have a centralized/distributed teams internally.

What’s ahead (20:00): Data science and AI will continue to grow and become important. People wanting to come into the industry to become data scientists should keep learning code, models, and algorithms. They should also try to put their technical expertise in the context of what does it mean for the business? What can or should the business do with the outcomes of the models? And they should be able to communicate the technology outcomes in the content of business outcomes.

Resources mentioned in this episode:

Podcast website: https://DataTransformersPodcast.Com

Data Transformers Podcast

Join Peggy and Ramesh as they explore the exciting world of Data Management, Data Analytics, Data Governance, Data Privacy, Data Security, Artificial Intelligence, Cloud Computing, Internet Of Things.

Data Science teams need people with multiple skill sets

‘

Career transition from Engineer to Finance to Data Science (02:30): Phl Bangayan started as an engineer at JPL but soon realized that needed a business background so went and got an MBA. After joining in Finance at Disney, Phil got an opportunity to shift into Marketing which led to an Analytics opportunity at Universal studios. From there, Teradata opened an opportunity to get into Data Science.

Analytics courses (07:00): As Phil started making decisions based on Analytics, he went to get formally educated on Data Analytics. MOOCs are a great way to learn but only 4% of the people who start a course actually complete it. Phil’s advice is that what works for him may not work for others regarding courses. In general, University offered courses are a better bet.

Move into Data Science (14:00): With an Analytics background, Phil now felt confident to move into Data Science. That is when Teradata opportunity opened up.

If no one understands a model, no good (18:00): It doesn’t matter how sophisticated your model is if no one understands it. And one of my pet peeves is when you do about 99% of the work, and then the work goes to waste because someone doesn’t agree with you, or because you forget to do that final part, which a lot of time is communicating it. And so I don’t believe that every data scientist has to be the best communicator, but I do believe that there has to be someone on the team who can take some analytic analysis and, and distill it to a point where an executive, like a CMO or CFO or president can, who doesn’t understand it can say, okay, what does this mean for me?

Data science needs complementary skill sets (21:00): In Data Science teams, define who you are, and don’t go off and compete against people who are more qualified than you are. So, for example, when I was going from marketing to data science, I knew that I did not want to compete against the 28 year old with a PhD in statistics. That person can write better code than me and can do better proofs than me and not a good, not a good competition. However, I bring a lot of experiences that someone who’s in that situation might not.

Last mile problem – Closing the deal (23:00): So if someone brings it 90% of the way through, I can help with that 10% and close the deal, or I can go ahead and make that work actionable. So that’s, so it doesn’t go to waste and, and no one wants to see their hard work go to waste. In fact, is one of my leadership pillars. I always tell my team that everyone plays a part it’s very easy when you’re the data scientist, to be able to say, well, hold it, I’ve got the hardest here because no one can do what I do.

Resources mentioned in this episode:

Podcast website: https://DataTransformersPodcast.Com

Episode Title : Data Science teams need people with multiple skill sets

Episode Summary: Data science career path doesn’t have to be purely technical. A data science team needs multiple skill sets. In this episode, Phil Bangayan, Principal Data Scientist at Teradata, talks about his career path from an electrical engineering background to MBA to Finance to Marketing and Data science. Phil talks about the need for the data science team to be able to communicate the outcomes of models in an understandable manner with CXOs. Phil also talks about the need for data science teams to have people with multiple skill sets.

Data Transformers Podcast

Join Peggy and Ramesh as they explore the exciting world of Data Management, Data Analytics, Data Governance, Data Privacy, Data Security, Artificial Intelligence, Cloud Computing, Internet Of Things.

What Does It Take To Be A Data Scientist?

‘

Episode Title : What Does It Take To Be A Data Scientist?

Episode Summary: How does one get to be a Vice President of AI at a global semiconductor leader? What professional journey can take from a Ph. D. to that influential role? Patrick takes the audience on a journey from upbringing in Germany, Malaysia, Philippines to education in the UK to an initial job at Los Alamos lab to the CEO of a startup to VP at Samsung. Later on Patrick talks about the skills and experience needed to be a data scientist and the emphasis on communication and business acumen to be a successful data scientist. The episode also discusses where leaders can learn from and what they should be doing continuously.

Journey from Ph. D. to VP (01:30): Patrick grew up in Malaysia and Philippines and went to University in the UK. Luckily Patrick got a job in Los Alamos laboratory. The research work needed a mathematics background so switched to math and started digging deeper into AI. Realized that you can’t apply AI while being a researcher.

AI research Versus Applied AI (05:20): Business case for AI is crucial. Lots of work in research institutions may not be commercially relevant and may not be viable. Having started an AI company, it became imperative to convince businesses that AI can pay.

Skills & Experience to be a Data Scientist (07:40): For a data scientist, programming knowledge in a language like Python is a must. Experience with frameworks like Tensorflow, Pytorch, Keras. (3) Pre-process and data prep (4) Statistical testing & probability analysis. So some mathematical & statistical knowledge along with data wrangling skills. On top of that, there are skills that are less common and more valuable is communication. Should be able to translate from domain english to business english. Talk & Translate.

Importance of Communication(12:00): Data scientists can somewhat easily learn mathematical and programming skills. They find the business communication the hardest.

Where can you learn? (13:30): Learn a great deal from my own department. Learn from my management. Learn from partners and customers. Learn from my own thinking like the Covid testing mentioned earlier. Source of most info personally is LinkedIn.

Future AI engagements (16:20): Currently pushing AutoML. Which models and which parameters. Most of the time people do it by trial and error. We are trying to automate this using AI. The other thing is distributed training. Training takes a long time but using multiple computers we can reduce the training time.

Reality of AI (20:00): AI is a big hype topic. But unlike other hype topics, AI is here to stay. Whatever your role is, get upto speed on AI. If you are a business person, think of ways to use AI.

Resources mentioned in this episode:

Podcast website: https://DataTransformersPodcast.Com

Data Transformers Podcast

Join Peggy and Ramesh as they explore the exciting world of Data Management, Data Analytics, Data Governance, Data Privacy, Data Security, Artificial Intelligence, Cloud Computing, Internet Of Things.

Developing Practical AI Applications – Patrick Bangert

‘

Episode Title : Developing Practical AI Applications – Patrick Bangert

Episode Summary: The episode focuses on developing practical applications using AI. Patrick Bangert, as the head of AI Engg and AI services at Samsung SDS, is in charge of including AI in almost all Samsung applications that are deployed on Samsung phones. If anyone is interested in learning the various phases of developing AI applications, this is the episode. Patrick discusses the avrious phases such as developing models, training the models, and deploying the models. The episode also goes over the proper characteristics of data for developing good ethical and explainable AI applications.

Types of AI practical applications (02:00) : Samsung SDS (Samsung Data Services) within Samsung. SDS is It services company; AI Engineering – Distributed Training, Auto ML, Algorithmic questions; AI Services –  Data science and AI projects for specific use cases. Natural language models for talking to your phone or Facial recognition for identifying. Textual recognition so you can answer that question.

Cost of training AI models (06:00) : Most of AI companies create AI products for other businesses and not for individuals. The models are not created for individuals as it is very expensive. Example is GPT3 model as the new state of the art language model from Google. The cost to train the model was $5 million. Only large companies can spend that amount of money.

AI Diagnostic models for Covid (10:00) (POTENTIAL FOR AUDIOGRAM). Ptarick’s team took the challenge of fast tracking Covid results using X-rays. You can take an X-ray of a chest to find accuracy of Covid tests. Hospital donated X-ray images (15,000 images); 6 doctors to assess X-rays; Lot of effort to label the data.

Relevant and Representative Data (14:50) – ANOTHER AUDIOGRAM – What problem are you trying to solve. What accuracy needs to be there? Raw dataset – 15,000 images. Data needs to be relevant. Data needs to be representative; Representation of the situation and Representation of the problem; (2) Label the data; Doctors may not be accurate all the time; May be accurate 80%. Labelling data is the domain expert’s job and not the scientist’s job

Ethical AI (21:00) – Patrick is on the board of AI Ethics journal; Some bad examples:  UK – AI was used to grade students; Facial recognition software was developed using caucasians but used on Africam people; Problems with Ethical AI are two-fold (1) Not a representative sample (2) Used on incorrect applications. Typically a data set problem

Explainable AI (23:30) – One of AI drawbacks is that it is a black box; The price for automation, reduction in costs is that it is not inherently explainable.

Why is it so difficult to explain AI? (26:20) – Neural network is about multiplying matrices with vectors.It is difficult to explain how this 1,000 coefficient equation came up with a prediction. Humans simplify assumptions. Neural networks will not simplify assumptions and compare details with details.

Resources mentioned in this episode:

Podcast website: https://DataTransformersPodcast.Com

Data Transformers Podcast

Join Peggy and Ramesh as they explore the exciting world of Data Management, Data Analytics, Data Governance, Data Privacy, Data Security, Artificial Intelligence, Cloud Computing, Internet Of Things.

Good Artificial Intelligence Governance is Good Business

‘

Episode Title : Good Artificial Intelligence Governance is good business

Episode Summary:

AI governance is about AI being explainable, transparent, and ethical. However, those three words mean different things to different organizations or functions within organizations, which results in slightly different definitions or descriptions of what AI governance is. 

David Van Bruwaene goes over his own professional journey which started with an undergrad degree in Philosophy. As he was traversing between philosophy and logic in his academia, David realized that data science and AI are emerging fields. After his education, David realized that there is very little focus on ethics and AI governance and started exploring that area. One thing led to another and quickly David was heading up an AI startup. After he sold the initial startup, David started an AI company focusing on ethical AI and AI governance.

A journey from Philosophy to Data Science (01:00): David traced his journey from an undergrad in Philosophy into AI. David kept getting better marks in logic. He came from a family of computer scientists. With a unique combination of philosophy and technology, David did lot of projects in AI, expert systems, and kept coding in LISP.

Does a background in linguistics give an edge in AI? (06:00): David believes that the background in philosophy and linguistics may have helped him be successful in building a company. Sure, people with a traditional comp science background will have an edge in coding etc. but having to deal with fewer resources in a startup gave David some advantage as he can extend beyond technology.

Academia to consulting to CEO (08:00): David ws brought in as a consultant to a company focusing on Ai while teaching at Waterloo. After the company decided to stop consumer focus, David was offered the CEO job as he knew the product very well. Even though the transition was not easy, it was fun to work with people in addition to working with products.

Transition from a CEO to found Fairly AI (12:00): After selling his previous company, David discussed with many people and wanted to focus on bringing AI to market but in a responsible way.

Influencers & Mentors (13:30): The work from Professor Floridi of Oxford Internet Institute and his team has been a great influence. Even movies like iRobot and Terminator influenced David and emphasized the importance of the balance between appeal of AI and the potential negative effects of AI. It is exciting to work on the human aspects of the technology as well.

What’s coming ahead in 2021 and beyond (17:00): There is potential legislation from the EU and some regulations from the US (from past stalled work) are expected. So some amount of algorithmic accountability will be expected. Even though the enormity of the task is daunting, humans working together on the policy front, technology, and regulations will have a better outcome. At least that is the hope.

Resources mentioned in this episode:

Learn more at http://www.DataTransformersPodcast.Com

Data Transformers Podcast dives into AI risk and AI ethics, with potential risks of AI we should probably be paying attention to now, if we want to develop the technology safely, ethically, and beneficially, while avoiding the dangers.

Subscribe to Data Transformers Podcast at: https://podcasts.apple.com/us/podcast/data-transformers-podcast/id1538437609

Twitter:

https://twitter.com/DataTransforme2

https://twitter.com/peggy_tsai

https://twitter.com/rkdontha1

LinkedIn:

https://www.linkedin.com/company/69241922

https://www.linkedin.com/in/peggy-tsai-data/

https://www.linkedin.com/in/rameshdontha/

Data Transformers Podcast

Join Peggy and Ramesh as they explore the exciting world of Data Management, Data Analytics, Data Governance, Data Privacy, Data Security, Artificial Intelligence, Cloud Computing, Internet Of Things.