If you happen to or your team are concerned with constructing ML solutions faster visit hf.co/support today!
👋 Welcome to Part 2 of our Director of Machine Learning Insights [Series]. Try Part 1 here.
Directors of Machine Learning have a novel seat on the AI table spanning the angle of assorted roles and responsibilities. Their wealthy knowledge of ML frameworks, engineering, architecture, real-world applications and problem-solving provides deep insights into the present state of ML. For instance, one director will note how using latest transformers speech technology decreased their team’s error rate by 30% and the way easy considering may also help save quite a bit of computational power.
Ever wonder what directors at Salesforce or ZoomInfo currently think concerning the state of Machine Learning? What their biggest challenges are? And what they’re most enthusiastic about? Well, you are about to search out out!
On this second SaaS focused installment, you’ll hear from a deep learning for healthcare textbook writer who also founded a non-profit for mentoring ML talent, a chess fanatic cybersecurity expert, an entrepreneur whose business was inspired by Barbie’s need to watch brand status after a lead recall, and a seasoned patent and academic paper writer who enjoys watching his 4 kids make the identical mistakes as his ML models.
🚀 Let’s meet some top Machine Learning Directors in SaaS and listen to what they should say about Machine Learning:
Omar Rahman
Background: Omar leads a team of Machine Learning and Data Engineers in leveraging ML for defensive security purposes as a part of the Cybersecurity team. Previously, Omar has led data science and machine learning engineering teams at Adobe and SAP specializing in bringing intelligent capabilities to marketing cloud and procurement applications. Omar holds a Master’s degree in Electrical Engineering from Arizona State University.
Fun Fact: Omar likes to play chess and volunteers his free time to guide and mentor graduate students in AI.
Salesforce: World’s #1 customer relationship management software.
1. How has ML made a positive impact on SaaS?
ML has benefited SaaS offerings in some ways.
a. Improving automation inside applications: For instance, a service ticket router using NLP (Natural Language Processing) to know the context of the service request and routing it to the suitable team throughout the organization.
b. Reduction in code complexity: Rules-based systems are inclined to get unwieldy as latest rules are added, thereby increasing maintenance costs. For instance, An ML-based language translation system is more accurate and robust with much fewer lines of code as in comparison with previous rules-based systems.
c. Higher forecasting ends in cost savings. Having the ability to forecast more accurately helps in reducing backorders in the provision chain in addition to cost savings because of a discount in storage costs.
2. What are the largest ML challenges inside SaaS?
a. Productizing ML applications require quite a bit greater than having a model. Having the ability to leverage the model for serving results, detecting and adapting to changes in statistics of information, etc. creates significant overhead in deploying and maintaining ML systems.
b. In most large organizations, data is commonly siloed and never well maintained leading to significant time spent in consolidating data, pre-processing, data cleansing activities, etc., thereby leading to a major amount of effort and time needed to create ML-based applications.
3. What’s a standard mistake you see people make attempting to integrate ML into SaaS?
Not focussing enough on the business context and the issue being solved, moderately attempting to use the newest and best algorithms and newly open-sourced libraries. Lots may be achieved by easy traditional ML techniques.
4. What excites you most concerning the way forward for ML?
Generalized artificial intelligence capabilities, if built and managed well, have the potential to remodel humanity in additional ways than one can imagine. My hope is that we’ll see great progress within the areas of healthcare and transportation. We already see the advantages of AI in radiology leading to significant savings in manpower thereby enabling humans to concentrate on more complex tasks. Self-driving cars and trucks are already transforming the transportation sector.
Cao (Danica) Xiao
Background: Cao (Danica) Xiao is the Senior Director and Head of Data Science and Machine Learning at Amplitude. Her team focuses on developing and deploying self-serving machine learning models and products based on multi-sourced user data to resolve critical business challenges regarding digital production analytics and optimization. Besides, she is a passionate machine learning researcher with over 95+ papers published in leading CS venues. She can also be a technology leader with extensive experience in machine learning roadmap creation, team constructing, and mentoring.
Prior to Amplitude, Cao (Danica) was the Global Head of Machine Learning within the Analytics Center of Excellence of IQVIA. Before that, she was a research staff member at IBM Research and research lead at MIT-IBM Watson AI Lab. She got her Ph.D. degree in machine learning from the University of Washington, Seattle. Recently, she also co-authored a textbook on deep learning for healthcare and founded a non-profit organization for mentoring machine learning talents.
Fun Fact: Cao is a cat-lover and is a mom to 2 cats: one Singapura girl and one British shorthair boy.
Amplitude: A cloud-based product-analytics platform that helps customers construct higher products.
1. How has ML made a positive impact on SaaS?
ML plays a game-changing role in turning massive noisy machine-generated or user-generated data into answers to every kind of business questions including personalization, prediction, suggestion, etc. It impacts a large spectrum of industry verticals via SaaS.
2. What are the largest ML challenges inside SaaS?
Lack of information for ML model training that covers a broader range of industry use cases. While being a general solution for all industry verticals, still have to work out easy methods to handle the vertical-specific needs arising from business, or domain shift issue that affects ML model quality.
3. What’s a standard mistake you see people make attempting to integrate ML right into a SaaS product?
Not giving users the pliability to include their business knowledge or other human aspects which might be critical to business success. For instance, for a self-serve product suggestion, it could be great if users could control the range of advisable products.
4. What excites you most concerning the way forward for ML?
ML has seen tremendous success. It also evolves rapidly to deal with the present limitations (e.g., lack of information, domain shift, incorporation of domain knowledge).
More ML technologies might be applied to resolve business or customer needs. For instance, interpretable ML for users to know and trust the ML model outputs; counterfactual prediction for users to estimate the choice consequence should they make a special business decision.
Raphael Cohen
Background: Raphael has a Ph.D. in the sector of understanding health records and genetics, has authored 20 academic papers and has 8 patents. Raphael can also be a frontrunner in Data Science and Research with a background in NLP, Speech, healthcare, sales, customer journeys, and IT.
Fun Fact: Raphael has 4 kids and enjoys seeing them learn and make the identical mistakes as a few of his ML models.
ZoomInfo: Intelligent sales and marketing technology backed by the world’s most comprehensive business database.
1. How has ML made a positive impact on SaaS
Machine Learning has facilitated the transcription of conversational data to assist people unlock latest insights and understandings. People can now easily view the things they talked about, summarized goals, takeaways, who spoke probably the most, who asked the most effective questions, what the subsequent steps are, and more. That is incredibly useful for a lot of interactions like email and video conferencing (that are more common now than ever).
With Chorus.ai we transcribe conversations as they’re being recorded in real-time. We use an algorithm called Wave2Vec to do that. 🤗 Hugging Face recently released their very own Wave2Vec version created for training that we derived loads of value from. This latest generation of transformers speech technology is incredibly powerful, it has decreased our error rate by 30%.
Once we transcribe a conversation we will look into the content – that is where NLP is available in and we rely heavily on Hugging Face Transformers to permit us to depict around 20 categories of topics inside recordings and emails; for instance, are we talking about pricing, signing a contract, next steps, all of those topics are sent through email or discussed and it’s easy to now extract that info without having to return through your entire conversations.
This helps make people a lot better at their jobs.
2. What are the largest ML challenges inside SaaS?
The largest challenge is knowing when to utilize ML.
What problems can we solve with ML and which shouldn’t we? Loads of times we now have a breakthrough with an ML model but a computationally lighter heuristic model is healthier suited to resolve the issue we now have.
That is where a robust AI strategy comes into play. —Understand how you would like your final product to work and at what efficiency.
We even have the query of easy methods to get the ML models you’ve built into production with a low environmental/computational footprint? Everyone seems to be combating this; easy methods to keep models in production in an efficient way without burning too many resources.
An incredible example of this was once we moved to the Wav2Vec framework, which required us to interrupt down our conversational audio into 15sec segments that get fed into this huge model. During this, we discovered that we were feeding the model loads of segments that were pure silence. That is common when someone doesn’t show up or one person is waiting for an additional to hitch a gathering.
Just by adding one other very light model to inform us when to not send the silent segments into this big complicated ML model, we’re able to save lots of loads of computational power/energy. That is an example of where engineers can consider other easier ways to hurry up and save on model production. There’s a possibility for more engineers to be savvier and higher optimize models without burning too many resources.
3. What’s a standard mistake you see people make attempting to integrate ML into SaaS?
Is my solution the neatest solution? Is there a greater technique to break this down and solve it more efficiently?
Once we began identifying speakers we went directly with an ML method and this wasn’t as accurate because the video conference provider data.
Since then we learned that the most effective technique to do that is to start out with the metadata of who speaks from the conference provider after which overlay that with a sensible embedding model. We lost precious time during this learning curve. We shouldn’t have used this massive ML solution if we stopped to know there are other data sources we must always put money into that can help us speed up more efficiently.
Think outside the box and don’t just take something someone built and think I even have an idea of easy methods to make this higher. Where can we be smarter by understanding the issue higher?
4. What excites you most concerning the way forward for ML?
I believe we’re in the course of one other revolution. For us, seeing our error rates drop by 30% by our Wave2Vec model was amazing. We had been working for years only getting 1% drops at every time after which inside 3 months’ time we saw such an enormous improvement and we all know that’s only the start.
In academia, larger and smarter things are happening. These pre-trained models are allowing us to do things we could never imagine before. This could be very exciting!
We’re also seeing loads of tech from NLP entering other domains like speech and vision and with the ability to power them.
One other thing I’m really enthusiastic about is generating models! We recently worked with an organization called Bria.ai and so they use these amazing GANs to create images. So you are taking a stock photo and you’ll be able to turn it into a special photo by saying “remove glasses”, “add glasses” or “add hair” and it does so perfectly. The concept is that we will use this to generate data. We are able to take images of individuals from meetings not smiling and we will make them smile to be able to construct a knowledge set for smile detection. This might be transformative. You may take 1 image and switch it into 100 images. This may also apply to speech generation which may very well be a robust application throughout the service industry.
Any final thoughts?
–It’s difficult to place models into production. Imagine data science teams need engineering embedded with them. Engineers ought to be a part of the AI team. This might be a very important structural pivot in the longer term.
Martin Ostrovsky
Background: Martin is enthusiastic about AI, ML, and NLP and is accountable for guiding the strategy and success of all Repustate products by leading the cross-functional team accountable for developing and improving them. He sets the strategy, roadmap, and have definition for Repustate’s Global Text Analytics API, Sentiment Evaluation, Deep Search, and Named Entity Recognition solutions. He has a Bachelor’s degree in Computer Science from York University and earned his Master of Business Administration from the Schulich School of Business.
Fun Fact: The primary application of ML I used was for Barbie toys. My professor at Schulich Business School mentioned that Barbie needed to watch their brand status because of a recall of the toys over concerns of excessive lead in them. Hiring people to manually undergo each social post and online article seemed just so inefficient and ineffective to me. So I proposed to create a machine learning algorithm that may monitor what people considered them from across all social media and online channels. The algorithm worked seamlessly. And that’s how I made a decision to call my company, Repustate – the “state” of your “repu”tation. 🤖
Repustate: A number one provider of text analytics services for enterprise corporations.
1. Favorite ML business application?
My favorite ML application is cybersecurity.
Cybersecurity stays probably the most critical part for any company (government or non-government) with regard to data. Machine Learning helps discover cyber threats, fight cyber-crime, including cyberbullying, and allows for a faster response to security breaches. ML algorithms quickly analyze the almost definitely vulnerabilities and potential malware and spyware applications based on user data. They will spot distortion in endpoint entry patterns and discover it as a possible data breach.
2. What’s your biggest ML challenge?
The largest ML challenge is audio to text transcription within the Arabic Language. There are quite a couple of systems that may decipher Arabic but they lack accuracy. Arabic is the official language of 26 countries and has 247 million native speakers and 29 million non-native speakers. It’s a fancy language with a wealthy vocabulary and plenty of dialects.
The sentiment mining tool must read data directly in Arabic if you happen to want accurate insights from Arabic text because otherwise nuances are lost in translations. Translating text to English or every other language can completely change the meaning of words in Arabic, including even the basis word. That’s why the algorithm must be trained on Arabic datasets and use a dedicated Arabic part-of-speech tagger. Due to these challenges, most corporations fail to supply accurate Arabic audio to text translation to this point.
3. What’s a standard mistake you see people make attempting to integrate ML?
Probably the most common mistake that corporations make while attempting to integrate ML is insufficient data of their training datasets. Most ML models cannot distinguish between good data and insufficient data. Due to this fact, training datasets are considered relevant and used as a precedent to find out the outcomes generally. This challenge isn’t limited to small- or medium-sized businesses; large enterprises have the identical challenge.
Regardless of what the ML processes are, corporations have to make sure that the training datasets are reliable and exhaustive for his or her desired consequence by incorporating a human element into the early stages of machine learning.
Nevertheless, corporations can create the required foundation for successful machine learning projects with an intensive review of accurate, comprehensive, and constant training data.
4. Where do you see ML having the largest impact in the subsequent 5-10 years?
In the subsequent 5-10 years, ML can have the largest impact on transforming the healthcare sector.
Networked hospitals and connected care:
With predictive care, command centers are all set to research clinical and site data to watch supply and demand across healthcare networks in real-time. With ML, healthcare professionals will have the ability to identify high-risk patients more quickly and efficiently, thus removing bottlenecks within the system. You may check the spread of contractible diseases faster, take higher measures to administer epidemics, discover at-risk patients more accurately, especially for genetic diseases, and more.
Higher staff and patient experiences:
Predictive healthcare networks are expected to cut back wait times, improve staff workflows, and tackle the ever-growing administrative burden. By learning from every patient, diagnosis, and procedure, ML is anticipated to create experiences that adapt to hospital staff in addition to the patient. This improves health outcomes and reduces clinician shortages and burnout while enabling the system to be financially sustainable.
🤗 Thanks for joining us on this second installment of ML Director Insights. Stay tuned for more insights from ML Directors in Finance, Healthcare and e-Commerce.
Big due to Omar Rahman, Cao (Danica) Xiao, Raphael Cohen, and Martin Ostrovsky for his or her good insights and participation on this piece. We stay up for watching each of your continued successes and might be cheering you on each step of the best way. 🎉
If you happen to or your team are concerned with accelerating your ML roadmap with Hugging Face Experts please visit hf.co/support to learn more.
