Women in Machine Learning Symposium 2022 Keynote
Hello everyone, we are live from the second Women in Machine Learning Symposium. My name is Rwana Karashkada, I’m the developer relations lead for Machine Learning Communities at Google and I will be a host for today’s event. Thank you all for joining us today. The Women in Machine Learning Symposium is an inclusive event for people to advance in machine learning and find a community of practitioners in the field. Regardless of your experience or expertise level, there’s a path for you at this event and afterwards. Last year we founded the Women in Machine Learning Program with a gold-off building and inclusive space for all intersections of diversity and to give a voice and platform to women. This year we focus on coming together to learn the latest machine learning tools and techniques, get a scoop on the newest ML products from Google and learn directly from several amazing women in the field. I would like to challenge you today to look at our event through the lenses of scale and impact. I invite you to think about how you can scale any lessons learned and how you can impact more people to learn about machine learning and to get involved in the field. We aim to inspire the next generation of machine learning practitioners, but more importantly, we want to empower everyone to learn and to use machine learning in their daily lives. Our curated lineup of speakers will share personal career stories, provide actionable feedback and advice on how to get started or how to advance in machine learning and will offer avenues to help you shape your career. Don’t forget to ask questions throughout the event. This is a great opportunity to connect directly with our machine learning team at Google who will be answering questions on chat. And just to give you a sneak peek, we have sessions on responsible AI and how to get started with diffusion models using KERES CV all the way to using deep learning in black hole weather forecasting. Without further ado, I would like to introduce you to our first keynote speaker. It is my honor to welcome Sherbani Roy, Senior Director at Google. Sherbani, the floor is yours. Thanks so much, Joanna. Hello to each of you from joining us around the world. It’s lovely to meet you. I’m Sherbani Roy, Senior Director in Google’s core machine learning group. I’m excited to help kick off today’s second women in Emelson, POSIUM. Today we’re here to celebrate women teaching everyone the latest and greatest information about machine learning. And to inspire each of you to be four small suppliers for ML innovation. What’s my path to machine learning, Gloria, you may ask? Well, first of all, I grew up outside of it Chicago and Atlanta as a nerdy kid who is never really good at hearing someone tell me no. I’ve always been more of a yes and kind of person with a healthy dose of pragmatism and intentionality. When I first started out on my professional journey, I didn’t have everything figured out and I did not take a clean linear path to where I am today. I’ve approached my life and career with an open-minded scientific mindset, experimenting, learning, and iterating in hypotheses about what I think I might or might not want to learn. I’ve prioritized consistent learning and trying to tackle the hardest problems I could find. I planted seeds, cultivated my skills, nurtured relationships, and always raised my hand to help out when I thought something could be interesting. This intentionality has helped my career blossom and accelerate as I grew into an engineering and product leader building applied ML products. And now, as a leader in delivering sustainable, responsible ML building blocks that make it simple to innovate at scale. Now I didn’t do this by myself. I’ve been fortunate to have a diverse set of mentors, including many strong women who have taught me important skills at critical points in my career. We know that machine learning is about using past data to predict the future, so let me walk you through some of these inputs that have been used to train the woman I am today. But the very start of my journey is my mom who established a foundation of curiosity and perseverance that has been critical every step of the way. Then about 25 years ago, on my first day of undergrad at Chicago, I met Alex, another woman who was also majoring in both physics and mathematics. We formed a study group with some other classmates that lasted all four years, but none of us would have made it without Alex. I remember one time, we were all gripping about a really difficult problem setting, quantum mechanics that had a single problem. I think many of you can probably relate to that horror story of that one problem set or exam that were deceptively short. She just gone downstairs to raid the vending machines in Harper Library and returned with a bag full of twistlers and cheetos she tossed on the table. She said something like, come on guys, this is just a problem set tough and up. Through the hundreds of problems that we worked on together across physics, math and stats, she never gave up and she would never let us give up. She even helped me work through the classic book, Teach Yourself C++ in 24 hours, which my professor gave me on my first day in his lab. Just because she thought it’d be interesting. With Alex, I didn’t just build foundational skills for my future ML journey, I also learned great in pragmatism. Then about 15 years ago, I was working as an analyst. I met Shannon. She ran a boot camp on stat in my first week in the San Francisco office, just as I’ve been dropped on a critical project with a tight deadline. I never used stat before and this isn’t a world before stack overflow or any real debugging help. Shannon teaching me and several other teammates these skills, not only helped me with short-term success, she also opened up my mind to a more applied career in data science instead of thinking it was reserved for academia. From Shannon, I didn’t just learn technical skills. I learned the importance of how to become a force multiplier for my team through intentional, consistent knowledge exchanges. And about eight years ago, when I started working in Alexa, my boss, Tony, taught hundreds of us how to develop deep ML AI product instinct. How many average customers knew what a life would be like, surrounded with voice assistance, smart cameras, and a world of multimodal interactions? How could we build the right future for customers when we were just inventing it? Tony mostly led by example, demonstrating deep respect for customers and was never afraid to ask tough questions like, should we actually build this? And even tougher questions like, should we stop building this? From Tony, I learned it’s not just about understanding the deep tech. You also have to remember who you are, building for, and how and why the magic you were building matters for those customers. About five years ago, when my team was testing out a new cutting-edge ML idea that we weren’t quite sure how we were going to land, one of my reports Anna, who’s now an engineering leader, gave me my first real introduction to this new framework called TensorFlow and helped truly accelerate my machine learning journey. She canceled her meetings and we spent a few hours whiteboarding. Anna didn’t just have the skills, she had enough confidence to know what she needed to focus on and what was best for the team to make sure we all knew what we needed to succeed together. Anna is a reminder that you can learn from anyone at every level. If it wasn’t for her competence and kind yet no nonsense approach, we may not have taken some incredibly important risks that resulted in some big wins. And today, you’ll meet a few other women who inspire me, including three of my teammates here at Google, Anita, Divya and Lou, and they’ll be speaking to the over 13,000 of you watching online. They’re some of my daily thought partners who bring deep expertise and vision to our team. And they’re focused on building incredible ML solutions like TensorFlow, Keras and Media Pipes, all part of our TensorFlow projects that help millions of developers build products for billions of people around the world. My goal for sharing these stories is to remind those of you watching that you should always feel empowered to lead, to speak up and ask questions, even if you don’t think you’re taking all the boxes. Their infinite seats at the table when it comes to building the future. And building the future is part of the reason I came to Google. I am here to be at the heart of accelerating the ML AI wave and to be a force multiplier for good. To be part of a company that not only can lead the technology to get us there, but also is fully committed to leading in an ethical, inclusive, and responsible way. Google is unique in that we provide end-to-end building blocks across the ML lifecycle that let anyone with any level of expertise build amazing innovation in ML, whether you’re taking your first steps as a learner or building massive language models. This includes everything across our TensorFlow projects from dealing with data, defining models across a variety of growing frameworks like TensorFlow, Keras and Jax, training models deploying them to various surfaces, including mobile, IoT and microcontrollers and web and all, and managing and operating systems that use them. And all of this is underpin by accelerated infrastructure for training and inference, as well as the tools that help you to be responsible across the board. We have ML offerings for everyone. And for everyone, it’s an inspiring time to get more involved as machine learning is involving at a rapid pace. In fact, we just saw in the last few months incredible advancements in generating images, videos and chat, and this doesn’t happen without empowerment of bidirectional communities to build and contribute to incredible advancements. And these communities themselves have evolved in incredible ways. So many of the specialized jobs that are critical today in machine learning didn’t even exist 10 years ago. Here at Google, we’re inspired by ML and the ML communities and building incredible offerings, like simple ML and Google Sheets, which allow you to pick up models from a simple drop down or media pipe, which powers high quality video experiences on Google Meet that improve how everyday families and colleagues are able to communicate and have fun together. Or with TensorFlow, which has enabled incredible advancements like breast cancer detection with our Google Health Memography AI research model being implemented in real world clinical practice. Today, you’ll learn more about these examples and more from each of our incredible speakers. They’re each deep experts in their area, and they are here to be helped be a force multiplier for the over 13,000 of you joining us from around the globe. Now I haven’t asked of each of you. As you watch today, I asked that you will think about what you will learn today that you can bring back into each others. How will you be a force multiplier for communities and around the world? All right. I think you’re all ready for Laurence Moroney, our AI Advocacy Leader here at Google, to take you through some details of incredible ML products being built by our communities. Thanks, Yourbani, and what a wonderful story. I’m Laurence, and I lead the AI Advocacy Team here at Google. What we strive for is that vision of openness, especially with a view to lowering the barriers of entry to everyone. And the work that the team is doing to bring AI and ML to the world through a comprehensive, free and open source ecosystem is really encouraging. And with that in mind, I’d like to first talk about how the MediaPipe team has been working to provide connective tissue across the data, modeling, and deployment columns. With MediaPipe Studio, this is particularly exciting for developers who want to use ML in their apps or sites, but don’t necessarily have that expertise in ML. The goal is to give you an easy way to create advanced ML solutions with easy to use, low code, and no code tooling. As well as simple to use APIs that integrate models into your apps or sites. Now models tend to be these binary blobs with tensors in and tensors out as the interface to your app. But with MediaPipe tasks, you have high level APIs that abstract this complexity, and they make the task of integrating models easier than ever. And of course, we’re delivering lots of new model types, making scenarios such as sentiment classification and hand gesture detection easier than ever. On the topic of hand gesture detection, we’re also excited to announce our first mobile competition on Kaggle, where we will focus on AI and ML helping the parents of deaf children learn how to use sign language. We’re putting the finishing touches on the data set now. So please sign up to get notified when we launch in the coming weeks. A message we constantly hear from ML developers is about how you want to make your models train and execute faster. One way to do this is with accelerated linear algebra or XLA, where ops in your model can get optimized for specific backend hardware such as GPUs or TPUs. To build the most open and helpful ecosystem and to make AI more accessible to developers, we’re working with leading organizations in the industry to collaborate on the open XLA project. This is an ecosystem of ML compiler and infrastructure technologies that lets you build your models using leading frameworks such as TensorFlow, PyTorch or Jax, and then optimize them to execute on diverse hardware. It helps you make the right choice of framework for your project. We’re continuing to work on it, so stay tuned. And continuing on in that spirit of openness and helpfulness, we’ve heard lots of feedback from developers and users about how complex ML can be. And there’s often many options and many moving parts, making it hard to even decide where you could get started. In particular, we’ve had feedback about recommender systems, where developers wanting to implement recommendations similar to those that you might see on e-commerce sites or YouTube. But with a variety of options, it can be really hard to ramp up. So I’m excited to announce the launch of our new developer site for recommenders on TensorFlow.org. It’s a one-stop shop for everything you need to understand how to use ML to build recommender systems, and it will get you up and running with TensorFlow recommenders. And guide you when you need to go that a little bit deeper to do things like optimizing with complimentary APIs such as TensorFlow ranking or Google scan. So check it out at this URL. But perhaps the most exciting way that I’ve seen of getting ML into the hands of more people is the amazing simple ML. And this puts ML into a tool that millions already use, and it gives you AI superpowers. But don’t take my word for it. I’d like to introduce my friend Dominique, and she’s somebody who’s already very familiar with superpowers. So take it away, Dominique. Wow, it’s a lot of exciting stuff coming up, Lawrence. And thanks so much for having me. Hello, I’m Dominique, and I play the superhero Iron Heart in the Marvel Cinematic Universe. I’m so honored to be here with you all today to give a shout out to all of you as the real heroes that know exactly how to harness the power of machine learning in your day-to-day lives. Just look at all the amazing things you’ve built with ML from helping to protect the great barrier reef to helping amputees. It’s incredible to hear all the ways you’re improving the world we live in. Even later on today, you’ll be hearing from ML engineers that work on black hole weather forecasting, and learn about the ways you can generate images from text. The possibilities with machine learning technology really seem endless. It’s clear to me that Hollywood and the media influence technology today. Just look at the movie back to the future. So many of the science fiction innovations foretold in that script are actually real today. And they were created by engineers like you. That’s why I’m so inspired by the opportunity to embody an engineer on screen. Because I know I’m representing all of you brilliant women in ML. I remember feeling strongly encouraged to apply to my performing arts high school and equally as strong about pursuing higher education at Cornell University. Once there, minoring in inequality studies, I pursued a degree in human development in social and personality development. I wasn’t doing anything related to acting with my major. So when I saw the artwork for Reary Williams or Iron Heart in my sophomore year, I was sufficiently blown away. For those of you who may not be huge Marvel fans like me, Reary Williams is an MIT student and an engineering prodigy with a super genius level intellect who builds the most advanced suit of armor since Iron Man. When I was approached about playing here, the first thing that came to mind was that evocative artwork I saw years back and how it made me feel seen and accepted. When I agreed to the role, I was most excited about the potential to give other people that might identify with Reary’s experience the same feelings of excitement and acceptance. The deeper I went trying to understand this character, the clearer I saw how many different flavors of STEM exist because of how many different kinds of people are interested and committed to the field. One of the most fulfilling things to tear down is the archetype of what a traditional scientist looks like. We could certainly think of a number of big brain geniuses in the superhero space. But seeing the imagery for Reary all those years ago, I had never seen a Marvel hero or engineer with my complexion with such a beautifully textured fro that was also a badass touting super genius intellect. That searing Reary was also most reflective of the truth of our world, which also includes the truth of women in STEM. They don’t all look the same. They don’t all come from the same place or share the same background. It’s really our differences that make us empowering. While my understanding of AI and machine learning is through a fictional lens, it is a prominent part of Iron Heart’s hero journey. A lot of her growth comes from trying to better build AI. And in a lot of ways, it represents a culmination and test of her genius. So it’s pretty cool to see how sophisticated AI is in the real world and how fast it changes. I know that today it is shifting into the mainstream and as regularly applies to our everyday lives. This has a lot to do with companies like Google making it easier and accessible for people to use. This example I’m going to share today is the new simple ML for Sheets. It’s an add-on for Google Sheets that helps make machine learning more accessible to all, especially folks without machine learning backgrounds. To demonstrate the power of simple ML for Sheets, we are going to use a fun example. Here, you will see that I have a data set with info on penguins from the Palmer archipelago in Antarctica. The data set has all kinds of traits like their built length, flipper length and body mass. And I will use simple ML to predict their penguin species from existing traits. Now let’s get started. First, I need to make sure I have the simple ML for Sheets add-on installed. To do that, you go to extensions, add-on, get add-ons, and you can search for simple ML for Sheets. Once you’ve added the add-on, it will live under the extensions menu and you’ll see it here on the bottom. Hover over it and a sidebar will show up. Then click show sidebar and it will load. Once it’s loaded, you can do a number of things. So what we want to do is predict the values in columns with empty cells. In this case, like I mentioned earlier, I want to know if the AI can correctly predict the species of penguin. Next, what we want to do is pick the columns that we will use to train our ML model. Each of these columns represents a variable like Billings or which island they were found. So let’s pick them all. And finally, I’m going to click predict. And after a few seconds, voila. You can see the model predicted the species of all penguins that were missing in the data set and with high confidence scores too. Just look at how easy and fun it can be to use ML. This was a playful example that I demonstrated, but this technology can be used in all types of scenarios. Imagine if you were a car repair shop owner. You can use historical records in your Google Sheets like car model type and length of repair time to predict how long a new repair might take. That would really help you plan out your business operations. Scientists can also benefit from ML in a lot of domains. For example, if you’re studying molecular aging, you can predict a person’s age based on DNA methylation data. And all of that could just be done in a few clicks. From small business owners to business analysts at large corporations, students to scientists, everyone familiar with Google Sheets can easily use and benefit from simple ML to improve in their field. Thanks for taking the time to listen to me and being a part of shifting the narrative in this space. Because if you are passionate about learning AI, building with AI or teaching AI, you truly are a superhero. I’d like to pass it back to Joanna, who will share the next part of today’s exciting agenda with you all. Take care. Thank you Dominique. We will now continue with the lining talks, followed by four workshops that will run in parallel, so make sure to choose the one that is the right fit for you. We’ll see each other again later today for our career panel with four amazing panelists.