Last week, I was happy to attend a conference on AI in San Francisco - AI Summit SF 2019.
There were some good talks at the conference.
Here is a summary of some learning, highlights and take-away from this conference.
1. ML/AI advantages and value are real: Companies are creating tangible value from 'institutionalizing' AI technology to their competitive advantage. The trend is seen to be spreading across the various large corporations much beyond the major players. There was a good presentation from American Express on how AI was used to create 90+% (!!!) efficiency in their Analytics team work. However, the Big Four are clear leaders in this space
2. Competitive Differentiation and Enterprise Adoption: KPMG made a good presentation on top trends in AI space and key takeaway here is that AI will be a key differentiation in competitive landscape of various industry verticals (Not exactly breaking news).
3. Stumbling blocks: The third topic is a little bit of dampener of sorts - many companies are failing to complete the full cycle of deploying the AI/ML work done in house. There was concensus that 60 to 70% of all ML models built are NOT deployed into production. My take on this is following - AI/ML output should fit into a process and make that process faster, better and smarter. This 'process-thinking' is not happening in many cases in companies. This needs to be thought through prior to Analytics/Model-building work and should have the support of the line-of-business process owners and leaders. This topic was also dealt with within the headline of 'ML deployment Maintenance & Management' which is a new and growing field and as per all the speakers,very different from traditional s/w maintenance (heuristic rules based s/w).
Another roadblock for the Data Science/ML teams here can be that Enterprise IT teams lack the skills/resources to deploy the ML work done and also primarily incentivized to work with the mindset of keeping Enterprise Systems up and running.
4. Auto ML: This is not an entirely new topic in software/platforms to assist in ML work. There are various software products and platforms that run through list of potential ML algorithms and choose 'best fit' algorithm for given data-set. These platforms are evolving to cover more of the data science work pipeline including deployment and model maintenance. These platforms don't come cheap
- H2O (they were not at the conference)
- IBM Watson Studio
- DataRobot
5. Data Collection Outsourcing: I learnt/discovered about an interesting outsourcing business model at the conference - Outsourced Data Collection/Digitization, Cleaning & Labeling for ML work.
These companies are providing for employment in lower cost regions of the world to enable ML. Go manual to go digital!
Below are some companies that had vendor booths at the conference that I interacted with.
- Appen
- Cloud Factory
- Indivillage
There were some good talks at the conference.
Here is a summary of some learning, highlights and take-away from this conference.
1. ML/AI advantages and value are real: Companies are creating tangible value from 'institutionalizing' AI technology to their competitive advantage. The trend is seen to be spreading across the various large corporations much beyond the major players. There was a good presentation from American Express on how AI was used to create 90+% (!!!) efficiency in their Analytics team work. However, the Big Four are clear leaders in this space
2. Competitive Differentiation and Enterprise Adoption: KPMG made a good presentation on top trends in AI space and key takeaway here is that AI will be a key differentiation in competitive landscape of various industry verticals (Not exactly breaking news).
3. Stumbling blocks: The third topic is a little bit of dampener of sorts - many companies are failing to complete the full cycle of deploying the AI/ML work done in house. There was concensus that 60 to 70% of all ML models built are NOT deployed into production. My take on this is following - AI/ML output should fit into a process and make that process faster, better and smarter. This 'process-thinking' is not happening in many cases in companies. This needs to be thought through prior to Analytics/Model-building work and should have the support of the line-of-business process owners and leaders. This topic was also dealt with within the headline of 'ML deployment Maintenance & Management' which is a new and growing field and as per all the speakers,very different from traditional s/w maintenance (heuristic rules based s/w).
Another roadblock for the Data Science/ML teams here can be that Enterprise IT teams lack the skills/resources to deploy the ML work done and also primarily incentivized to work with the mindset of keeping Enterprise Systems up and running.
4. Auto ML: This is not an entirely new topic in software/platforms to assist in ML work. There are various software products and platforms that run through list of potential ML algorithms and choose 'best fit' algorithm for given data-set. These platforms are evolving to cover more of the data science work pipeline including deployment and model maintenance. These platforms don't come cheap
- H2O (they were not at the conference)
- IBM Watson Studio
- DataRobot
5. Data Collection Outsourcing: I learnt/discovered about an interesting outsourcing business model at the conference - Outsourced Data Collection/Digitization, Cleaning & Labeling for ML work.
These companies are providing for employment in lower cost regions of the world to enable ML. Go manual to go digital!
Below are some companies that had vendor booths at the conference that I interacted with.
- Appen
- Cloud Factory
- Indivillage
Comments
Post a Comment