Google’s first ever summit for builders utilizing AI on the consumer aspect
On 18th October, 2024, Google hosted the very first Web AI Summit to deliver collectively high minds from all over the world working with machine studying fashions client-side in the web browser. This implies after the preliminary web page load, all of those options might work totally offline on the consumer system, permitting the customers to learn from low latency inference, lower costs, and privacy.
Our lineup included presenters not solely from Google’s groups corresponding to Chrome and MediaPipe, but additionally energetic 3P representatives within the area corresponding to Intel, Hugging Face, Microsoft, LangChain and past. From shopper packaged items detection to healthcare options – talks lined a variety of industries and topic areas displaying simply how far Internet AI can attain.
Learn on for extra particulars or view the YouTube playlist to catch up right away and see the talks for yourself!
Inaugural Internet AI Summit highlights
We had over 1,100 registrations from folks spanning 22 nations, 59 cities, and 179 totally different Google places of work becoming a member of us for this historic occasion with a full home the entire day – it was great to see how engaged everyone was for the talks.
We had a mix of software program engineers, enterprise resolution makers, and govt management within the viewers, making a productive synergy between technical experience and strategic planning.
Internet AI Summit audio system and periods
Our skilled audio system shared invaluable insights to equip Javascript builders with data on subtle and complicated AI-powered options which are turning into the trade customary to fulfill purchasers’ calls for. Examine all of the talks under or view them your self on a espresso break.
Welcome to Internet AI Summit 2024
Jason Mayes – Web AI Lead, Google
Jim Bankoski – VP Engineering, Chrome, Google
An outline for the state of Internet AI in 2024 and why the Internet AI Summit was created. See what’s doable with machine learning on-device, in addition to the place it’s heading, to get the 101 earlier than you watch the opposite talks on this sequence. This speak is appropriate for everybody and covers topic areas corresponding to generative AI, LLMs, diffusion fashions, WebGPU, WebAssembly, and rising APIs like WebNN together with examples from trade which are already utilizing Internet AI at the moment.
Transformers.js: State-of-the-art Machine Studying for the Internet
Joshua Lochner – ML Engineer (Transformers.js), Hugging Face
Find out about Transformers.js, an thrilling new JavaScript library that empowers builders to construct never-before-seen net functions. It’s designed to be functionally equal to Hugging Face’s Python transformers library and helps over 120 architectures throughout a various set of duties and modalities. Customers can select from over 1,000 pretrained fashions or convert their very own to run domestically within the browser, providing privacy-preserving, low-latency, and scalable machine studying. The most recent addition of WebGPU help permits highly-performant execution of fashions by using trendy GPU capabilities straight within the browser.
The Internet Neural Community (WebNN) API: The place we’re and What’s Subsequent
Rob Kochman – Group Product Manager (Chrome), Google
Rafael Cintron – Principle Software Design Engineer, Microsoft
Superior net applied sciences like WebAssembly and WebGPU have lately introduced actual AI capabilities to the browser. The proposed Internet Neural Community (WebNN) API goals to construct on that momentum, enabling AI workloads to run sooner and extra effectively on quite a lot of units, together with units with AI accelerator {hardware} (NPUs), all based mostly on net requirements. This session will begin with a quick overview of WebNN, then describe current developments, together with API form, system help, framework help, and browser implementations. We’ll additionally describe the plan ahead, as we work to get suggestions from the group.
Internet AI on AI PC
Intel showcased WebNN, an rising unified W3C net customary API for on-device net ML acceleration throughout consumer AI execution engines: CPU, GPU, and NPU. At present in Developer Preview on Chrome and Edge browsers and built-in in common ML frameworks (e.g., ONNX Runtime Internet), WebNN delivers “near-native” efficiency and energy traits. We are going to present thrilling WebNN demos and adoption previews that deliver a brand new class of experiences to the net.
ml5.js – Pleasant Machine Studying for the Internet
Yu Lee – Resident Researcher, ML5.js, NYU
Aidan Nelson – Visiting Faculty, ML5.js, NYU
This speak targeted on ml5.js – an open supply library constructed on high of TensorFlow.js with a objective of creating machine studying approachable for a broad viewers of artists, artistic coders, and college students. This undertaking has been constructed as a collaborative effort at NYU’s ITP Program, drawing inspiration from Processing and the p5.js undertaking’s concentrate on making coding accessible and inclusive. ml5.js goals to increase this mission to the area of machine studying, bridging the hole between the technical complexity of machine studying and the creativity of novices and artists.
WebLLM: A Excessive-Efficiency In-Browser LLM Inference Engine
Charlie Ruan – student researcher, CMU
This speak lined WebLLM, a high-performance in-browser LLM inference engine. WebLLM permits constructing AI-enabled net apps which are quick (native GPU acceleration through WebGPU), personal (100% client-side computation), and handy (zero surroundings setup). For builders, WebLLM options an OpenAI-API type interface for standardized integration, helps chat functions and environment friendly structured JSON technology, and affords built-in help for Internet/Service Staff to separate backend executions from the UI circulate. On this speak, we are going to discover WebLLM’s key options, total structure, and the way builders can construct AI-enabled net functions with it.
State Is not All You Want, However It Helps: Constructing Higher LLM Apps within the Browser
Jacob Lee – Founding Engineer, LangChain
Thrilling new advances from tasks like WebLLM, Transformers.js, and Chrome AI have introduced native LLMs nearer than ever to anybody with a browser. This has immense potential to increase the frontiers of net growth, however these small fashions are extra restricted than state-of-the-art hosted fashions and require extra cautious concerns round design and prompting.
This speak targeted on addressing these constraints by masking strategies for implementing sensible apps that take advantage of small fashions utilizing the highly effective toolkit supplied by LangGraph.js, a brand new framework for orchestrating stateful LLM apps.
Visible Blocks: Visible Prototyping of AI Pipelines
Ruofei Du – Interactive Perception & Graphics Lead, Google
Visual Blocks for ML is a visible programming platform that empowers speedy AI and multimedia prototyping. On this speak, we are going to showcase how you can construct interactive AI pipelines, carry out interactive knowledge augmentation, and take a look at pipelines with stay knowledge utilizing easy drag-and-drop actions. We can even spotlight a spread of community-contributed pipelines and customized nodes demonstrating various functions in interactive graphics, massive language mannequin chains, laptop imaginative and prescient, and multi-modal options. Lastly, we encourage all Internet AI practitioners to contribute their very own ML pipelines and customized nodes, additional enriching the shared platform and provoking progressive use circumstances.
Exploring different interactions in JavaScript
Charlie Gerard – Senior Research Engineer, CrowdStrike
The most recent developments in AI have primarily targeted on massive language fashions and new methods of making and consuming content material. Nevertheless, AI additionally affords the chance to rethink the way in which we work together with interfaces. Utilizing JavaScript and fashions targeted on physique monitoring or audio classification, net builders have a novel alternative to experiment with different interactions to create extra progressive net experiences.
Overview of Chrome built-in AI
Kenji Baheux – Product Manager, Chrome, Google
Sharing what we have been as much as in Chrome for built-in AI, what we have realized, and what’s subsequent. We’ll discuss how we see the Immediate API, our standing for high-level activity APIs corresponding to summarization, write / rewrite, learnings from the early preview program, and the place we’re going from right here.
Internet AI in Trade: How TensorFlow.js has pushed what you see on the grocery store cabinets
Hugo Zanini – Technical Project Lead, Nubank
This speak showcased how one of many high 10 largest shopper packaged items (CPG) firms on the planet utilized Internet AI to increase its in-store commerce advertising technique in Brazil and the way it advanced into an open-source undertaking that has been helpful to different firms within the trade.
Classes realized from being buyer zero of Chrome’s built-in APIs
Thomas Steiner – Developer Relations Engineer, Chrome, Google
On this speak, Thomas summarized a number of the issues Developer Relations has learnt of their position as buyer zero of Chrome’s built-in APIs. Utilizing an instance of an AI-powered synonym finder app, he’ll present how you can work with the Immediate API specializing in facets from tweaking the immediate, to reliably parsing the output, to optimizing the app for optimum efficiency.
The Way forward for AI is Now: Actual-life Case Research for on Shopper-side AI Adoption in Internet Apps
Yuriko Hirota – Partner Solutions Engineer, Google
Kazunari Hara – Developer Expert, CyberAgent
This lightning speak reveals the sensible energy of client-side AI not only for the sake of utilizing AI, however for enhancing person experiences. The speak took a deep dive right into a real-world case examine featured in Google I/O 2024, showcasing how CyberAgent, the powerhouse behind certainly one of Japan’s high weblog providers, plans to leverage the magic of client-side AI to empower customers with easy weblog title technology. Be part of us to find out how CyberAgent maximized the potential of client-side AI by way of progressive use case design and a user-centric method.
Why are Internet Extensions improbable for AI?
David Li – Product Manager, Chrome, Google
On this speak we are going to showcase the potential of AI and Chrome Extensions. Chrome Extensions help you management the browser, observe net content material, and add your personal UI. When mixed, AI and Chrome Extensions could make the searching expertise actually useful and extra productive. This speak will give an summary on how extensions on the WebStore are utilizing AI at the moment and the place we see the most important potential.
Past the Banner: The Energy of Internet AI to Personalize Paid Media
Uncover how Internet AI is revolutionizing personalised paid media by introducing groundbreaking digital try-on advertisements throughout each display.
MediaPipe Internet: Bringing cross-platform AI tech to the browser
Tyler Mullen – Staff Software Engineer, Mediapipe, Google
Find out about MediaPipe’s cross-platform method to constructing AI pipelines and bringing them to the browser. We’ll spotlight a number of the advantages of our methodology and discuss a couple of of the foremost merchandise we assist energy (like Google Meet). Then we’ll cowl our newest technological developments and developer APIs. These choices embrace options for conventional machine studying duties like picture segmentation, in addition to generative AI duties like LLM inference. Lastly, we are going to give a sneak peak into the long run with some thrilling demos!
Remodeling entry to healthcare by way of WebAI
Chris Slee – CTO, Include Health
IncludeHealth, a digital bodily remedy supplier, harnesses the facility of WebAI to interrupt down logistical and financial obstacles, permitting sufferers to obtain personalised, measured care anyplace, any time, and on any system
ML Coaching on the Internet: Constructing Easy ML for Google Sheets
Richard Stotz – Software Engineer, Core ML, Google
Learn the way we constructed Easy ML for Sheets, a free Google Sheets add-on for ML and AI. Easy ML for Sheets makes use of on-device Machine Studying powered by WebAssembly, Javascript and Chrome’s new built-in AI to unlock superior Machine Studying duties for all customers. This speak highlights the instruments we used to efficiently deliver Easy ML for Sheets to market and the way our workforce’s open supply libraries assist builders obtain their very own ML successes on the net.
Thanks to everybody!
This occasion wouldn’t be doable with out the quite a few individuals concerned within the creation and working of the occasion. We want thank our 3 occasion creators, Jason Mayes, Jenna Zheng, and Marcus Chang for placing on the occasion and naturally an enormous thanks to all of our presenters listed above, together with our helpers and assistants on the day, and our AV groups who ensured the run of present was easily recorded in your viewing pleasure after the present.
Need to attend the following Internet AI Summit?
For those who missed the occasion this time round, catch up through the movies above, and make sure to subscribe to our public Web AI Newsletter to learn after we subsequent go stay!
