Introduce yourself! #35
Replies: 32 comments 29 replies
-
|
Hello, my name is Jeremy Ellis. You can find me on GitHub at https://github.com/hpssjellis and on LinkedIn at https://www.linkedin.com/in/jeremy-ellis-4237a9bb/ I've been creating neural networks for nearly 30 years and teaching high school machine learning for 8 years. Where I lack in book writing skills, I excel in simplifying ML concepts for teaching. In the cs249r_book, several chapters have discussed the security advantages of tinyML. You can explore these insights in the following links: embedded_sys.qmd Line 61 in d9c33af embedded_ml.qmd Lines 154 to 156 in d9c33af and also Line 204 in d9c33af However, it's worth noting that these chapters don't seem to mention that most tinyML models are created on cloud platforms. This inherently poses a security risk, as data is transferred during the training process, and there are questions about the handling of that data while it resides on the cloud. My particular area of interest revolves around streamlining the process of efficiently developing client-side machine learning models for tinyML. The question at hand is whether the construction of tinyML models is as secure as we assume, and if there are alternative approaches worth considering? |
Beta Was this translation helpful? Give feedback.
-
|
Hey @hpssjellis! Sorry that this message missed my inbox, I have been overwhelmed. Thanks for jotting down the points. I agree there is room for improvement. As we are working through the chapters, these points are covered in the later chapters, or at least I'd like to think so. Please continue to look through the other chapters, as they become available. Feel free to make suggestions or edits, as you see fit and issue a PR if you'd like to. We are open for contributions! Some of the specific things about security etc. will be covered in the on-device learning and security/privacy chapters 👍 |
Beta Was this translation helpful? Give feedback.
-
|
Hi all, Jared Ping here from South Africa. Group Technical Manager at Digital Matter, where we have a huge focus on the energy efficiency of battery-powered tracking devices in a range of applications (TinyML included!). Currently pursuing a part-time MSc(Eng) Electrical Engineering at the University of the Witwatersrand, specialising in operational optimisation of TinyML systems. Very excited to see the TinyML community grow, and this book is a great resource for introducing peers to it! |
Beta Was this translation helpful? Give feedback.
-
|
Hi @JaredP94, this is fantastic to read. I'm sorry that the message is long overdue. Your work at Digital Matter sounds like a perfect real-world application of TinyML. I love that you’re blending hands-on engineering with academic work at Wits. Really glad to have you in the community, and I hope the book becomes a helpful resource as you push forward on energy-efficient systems. We have been updating the kit support, which I am sure you must be aware of here: https://www.mlsysbook.ai/contents/labs/kits.html |
Beta Was this translation helpful? Give feedback.
-
|
Hello everyone, My name is Ethan Wang, a Master's student in Software Engineering at North China Electric Power University. My journey in the tech world is driven by a profound passion for operating systems (OS) and the burgeoning field of Tiny Machine Learning (TinyML). The challenge of maximizing the capabilities of limited resources to achieve seemingly limitless tasks fascinates me. Currently, I am interning at Megatronix, where I am honing my skills in system stability, a crucial aspect of software engineering that aligns closely with my interests in OS and TinyML. My Main Interests:
Future Aspirations:My immediate goal is to deepen my understanding and expertise in OS and TinyML through practical experiences and academic research. However, looking ahead, I am also setting my sights on opportunities to study and work abroad. I believe that pursuing further education and professional experiences abroad will not only broaden my horizons but also provide me with a unique perspective on global technological challenges and innovations. Contributions to the Community:
Additional Links:
I am looking forward to engaging with this community, learning from you all, and contributing where I can. Thank you for welcoming me into this vibrant space! |
Beta Was this translation helpful? Give feedback.
-
|
Hi @artwalker, welcome to the community, and pardon the slow response; I didn't realize these messages were coming through as they were getting filtered on my end. Your passion for operating systems and TinyML really comes through here. I love how you’ve framed TinyML as doing “infinite things with limited resources” 👍 That captures the heart of why this field is so exciting. I’m glad you’ve joined us, and I look forward to seeing how your work at Megatronix and your future research unfold. |
Beta Was this translation helpful? Give feedback.
-
|
Nice to see that this is an active repo and work in progress with commits as latest as yesterday! |
Beta Was this translation helpful? Give feedback.
-
|
Welcome @Seshu-R, please feel free to share any thoughts and feedback you might have. We have some exciting new updates coming soon! I have been working on this for a long time. Regarding the exercises and labs, I initially had some exercises and apps included, but since the beginning of the year, we've been working hard to update them so they are more cohesive: https://www.mlsysbook.ai/contents/labs/kits.html and coming soon! https://github.com/mlsysbook/TinyTorch. Please don't distribute that link broadly as it's still very much a work-in-progress. The only reason it's even live is because I'm trying to get some feedback. |
Beta Was this translation helpful? Give feedback.
-
|
Hello everyone, You can find me on:- |
Beta Was this translation helpful? Give feedback.
-
|
If you're interested in robotics, I strongly encourage you to engage in practical deployment scenarios of AI. Robotics and AI require embedded machine learning technologies and capabilities. Both these domains are computationally very intensive and they're also severely power constrained, therefore HAI technologies are going to be quite critical for them. |
Beta Was this translation helpful? Give feedback.
-
|
Hello Professor Reddi and everyone! 👋. I'm Hardik Agarwal, an undergraduate student at the Indian Institute of Technology, Delhi. My interest in ML ignited when I saw how effective it can be in tackling real-world problems with surprising efficiency and precision. Over time, I began to explore many aspects of machine learning, and I am presently most interested in TinyML. The idea of deploying machine learning models on resource-constrained devices and optimizing them for edge applications is both daunting and interesting to me. In the future, I hope to contribute by developing more efficient AI models and robotic systems that can operate seamlessly in real time and with minimal energy consumption. I’m really passionate about the intersection of machine learning, robotics, and optimization, and I see this as the key to unlocking new possibilities in various industries. I've been reading your Machine Learning Systems book, and I must say that it's quite informative and useful. Your focus on the practical aspects of ML deployment and system optimization, which are frequently disregarded in other resources, fills a critical need. Thank you for creating such a useful resource! I'm looking forward to networking with everyone and learning more about ML systems. I’m eager to learn from your expertise and insights as well as from my fellow students who share my passion for machine learning and robotics. I will be happy to connect here: |
Beta Was this translation helpful? Give feedback.
-
|
Thank you for reaching out! I'm always excited to see students engaging with this resource. You are the reason I feel inspired to keep developing it. I remember being deeply motivated to study computer architecture as a student because of Hennessy and Patterson's textbook on computer organization and advanced computer architecture. If you are interested in TinyML -- checkout tinyML on HarvardX. There are many other resources I can recommend, but this is a great starting point! You can take it for free and if you need a kit for Course 3 let me know. |
Beta Was this translation helpful? Give feedback.
-
|
Hello Everyone, 👋 I'm Poojak Patel, a student at Illinois Mathematics and Science Academy. Since a young age, I have wondered how cars operate, which has grown into a broader interest in how machine learning could be used to enhance autonomous systems. Currently, I am interested in how we can create autonomous systems using machine learning that can help robots learn in real time. The idea that machines can both react to what's happening around them while learning from their experiences just like how a person would is fascinating. Also, after reading more about TinyML in Professor Reddi’s Machine Learning Systems book, I have been intrigued about its potential as it allows machine learning models to run on small devices with low power. I think, just like machine learning, TinyML can lead to more innovative solutions in the future. I am excited about the possibilities that this kind of technology could bring and how it would be able to handle current existing challenges, making a difference in people's lives. I would love to connect with others who are also interested in machine learning, robotics, or any related topics. Feel free to reach out to me by email (patel.poojak@gmail.com) and see my past projects on GitHub (github.com/PatelPoojak). Looking forward to engaging with all of you in this community! Best Regards, |
Beta Was this translation helpful? Give feedback.
-
|
Hi Poojak 👋 Nice interests. Embodied AI is expected to be a significant topic in the coming years. I believe the next major challenge will involve how language models, like ChatGPT, can begin interacting with physical sensors in the real world. This interaction will provide them with a wealth of new information and data to process. Therefore, I genuinely think we are only at the tip of the iceberg. |
Beta Was this translation helpful? Give feedback.
-
|
Hi everyone! I’m Mimi, and I currently work in helpdesk but am aiming to transition into software development. I'm starting with front-end development through a Meta course and am really enjoying learning the basics. I'm eager to expand my skills and learn from this community, so any tips or advice would be greatly appreciated as I make this career shift! I would love to connect! |
Beta Was this translation helpful? Give feedback.
-
|
Hello and welcome! A general word of advice -- it's important to complement your knowledge with AI tools, as they will play a major role as a copilot in assisting us with programming tasks. Therefore, becoming proficient with these tools while still learning the fundamentals (which is equally important) will be essential for all of us moving forward. |
Beta Was this translation helpful? Give feedback.
-
|
Thank you for the warm welcome and the advice! I completely agree that AI tools are becoming an essential part of programming. I'll definitely look into learning more about them while continuing to strengthen my fundamentals. It’s exciting to see how they can complement our skills and enhance our efficiency! Any suggestions or resources I can reference to start learning about AI tools? |
Beta Was this translation helpful? Give feedback.
-
|
Since your last message, as I'm sure you'd agree, so much has changed in the world with a lot of agentic work. I don't know if the following link will be useful to you or not, but nonetheless I feel like it might spark some ideas: https://harvard-edge.github.io/cs249r_fall2025/ |
Beta Was this translation helpful? Give feedback.
-
|
Hi all! My name's Ryan. I'm currently working at Apple doing cloud/backend engineering with a focus now on testing and system integration. No practical ML experience at this time. Working at Apple has inspired a data-privacy-centric view in me. That, combined with an increasing personal desire to spend less time on socially connected, networked devices has spurred an idea in me which involves using federated machine learning on a dis-connected personal device. My desire is to create a simple AI assistant that is vocally interacted with (no screen) and dis-connected from the internet (on-device learning with data privacy). Soon after I had this thought, I came across this book and thought it would be a great way to dive into ML, especially after hearing about TinyML and its relevance to the on-device learning I want to do. I've never worked with hardware, so this will be a fun and new challenge. Fun fact: Another passion project of mine is electronic music production. |
Beta Was this translation helpful? Give feedback.
-
|
Hi @orpheus -- Ryan, great to have you here. Your cloud and systems background will actually give you a solid foundation as you step into ML and hardware. The shift toward privacy-centric, local intelligence is a big theme in the field right now. Have you checked out Moonshine AI -- it is by Pete Warden. A voice interface on a fully disconnected device is both technically challenging and beautifully aligned with where the ecosystem is heading. Check it out here: https://petewarden.com/2024/10/21/introducing-moonshine-the-new-state-of-the-art-for-speech-to-text/ |
Beta Was this translation helpful? Give feedback.
-
|
Hi everyone, I’m Danjie, an undergraduate at University of Toronto. My first exposure to ML systems came when I was trying to integrate Apple’s Neural Engine with a personal project. I managed to get BERT running on my M1 laptop (with some reshaping and some other steps that I don’t really remember). The results were incredible, the Neural Engine achieved seven times the throughput of the M1’s GPU while consuming just 5 watts of power! On top of that, I have a close friend working at Annapurna Labs in Toronto on AWS’s Inferentia service. He told me they were working on getting Llama3 405b to run on Inferentia chips. It’s exciting to see how ASICs are rapidly gaining traction in the industry and generating millions in revenue for AWS. I’m particularly passionate about the co-design of hardware and software for AI acceleration. My background includes experience in computer architecture, assembly language, and high-level machine learning with PyTorch. This background allows me to see that there’s immense potential to improve efficiency across all kinds of applications, though I’ve often struggled to find resources on this topic. Thanks so much for the course! All the best, |
Beta Was this translation helpful? Give feedback.
-
|
That story about the Neural Engine is fascinating, @DanjieTang. Apple has made many quiet but impactful moves in this area. If you’re already considering hardware–software co-design, you’re looking at one of the most strategic levers in ML systems right now. Have you thought about exploring or even benchmarking how those accelerators compare to something like Inferentia or Coral? A personal project like that could provide valuable insights for the community. |
Beta Was this translation helpful? Give feedback.
-
|
Hi my name Is Tabina Hussain, I am a current high school student at Nikola Tesla STEM High School with a focus pursuing Mechanical & AI engineering. My current specialized focus is making cars more efficient and environmentally friendly. To get closer to this goal I’ve started researching energy efficiency aerodynamics and materials, like for example improved sensors that use AI to calculate the most optimizing solutions. This could push automotive technology forward. Making solutions to these types of issues encourages me to participate in the future of transportation. Deeper in that topic, something that intrigues me that I found in the MLSYS Book was the section on AI training due to its incorporation in EV sensors. I hope to be a part of the evolution of this principle. So excited to be an apart of this community, feel free to reach out to me at tabinazaheen@gmail.com for anything Mechanical or AI related. Always ready for involvement! Many Thanks, |
Beta Was this translation helpful? Give feedback.
-
|
Hi @KatsuKorr -- I love how you’re already connecting AI with aerodynamics and energy efficiency. That’s exactly the kind of systems-level thinking that pushes fields like transportation forward. One of the most exciting frontiers in this space is how embedded sensors and lightweight AI models can work together to make real-time decisions on the edge. If you ever decide to prototype or explore a specific problem (like optimizing drag in real time), the community here would be a great place to share that process. Even small experiments can lead to big insights. For starters, check out these kits - https://www.mlsysbook.ai/contents/labs/kits.html, and we are putting more together and welcome ideas and suggestions. |
Beta Was this translation helpful? Give feedback.
-
|
Hi everyone! I'm Avinash Krishna, in the final year of my Master of Technology in Electrical and Computer Engineering at IIT Jodhpur, India. My journey in technology has taken me through diverse areas, from machine learning and AI, cloud computing and network security. I've worked on research projects at IIT Delhi, IIT Dharwad, and IIT Jodhpur, applying AI/ML, 6G communication, 5G network slicing, and security systems. My interest in ML systems began when I saw how AI can revolutionize real-world applications—AI, ML, and CV for autonomus system, Industry 4.0, improving the reliability of 6G networks, and enhancing security with fewsshot learning based concealed object detection. The ability of ML models to find patterns, make predictions, and automate complex decisions fascinates me. Few-shot learning in AI/ML is something that truly excites me! I've worked on few-shot learning-based concealed object detection using YOLO, ResNet, and GANs in sub-terahertz security imaging. The potential of ML systems to learn from minimal data and adapt quickly to new tasks is incredibly powerful, especially for security, healthcare, and automation applications. I'm looking forward to talking with others. |
Beta Was this translation helpful? Give feedback.
-
|
Hi @avinashverse, I really appreciate you bringing up ML systems. The term is used in many ways, and I find that clarifying what we mean by it leads to better conversations. When I talk about ML systems, I think of everything surrounding a model, including data pipelines, runtime, serving, hardware, and components that enable scale. I'm curious what it means in your work, since your usage seems closer to modeling. I'm just being a bit particular because I care a lot about how we use these words. See our textbook definition here: https://www.mlsysbook.ai/contents/core/introduction/introduction.html#sec-introduction-defining-ml-systems-bf7d Also, I want to say it’s great to see someone already working on few-shot learning in applied security contexts. That is a real and challenging problem space, and I look forward to seeing how you bring your perspective into this community. |
Beta Was this translation helpful? Give feedback.
-
|
Hi everyone! I’m Tadiwos, an undergrad at KAIST, double majoring in EE and CS. I’m interested in ML systems, particularly in efficient AI—accelerating inference and hardware-software co-design. My research focuses on systems for ML and AI accelerator architectures, with a focus on optimizing inference performance. My interest in ML systems was sparked by the realization that while AI models continue to grow in complexity, their impact hinges on the efficiency of the underlying hardware and system optimizations. I've worked on inference acceleration, GPU performance analysis, and algorithm-hardware co-design strategies, particularly in areas like tensor computation efficiency and distributed inference. |
Beta Was this translation helpful? Give feedback.
-
|
Hi @TadiwosAbay This is a really sharp way of framing the problem. I like that you are focusing on efficiency as the real enabler of impact, not just bigger models. The fact that you are already working on accelerator architectures and co-design strategies puts you right in the middle of some of the most important questions in this field. I am curious about how you think about the trade-off between model and system design when working on inference optimization. That space is where a lot of the most interesting ideas live. |
Beta Was this translation helpful? Give feedback.
-
|
I am Aryaa Gargatti, a Bachelor of Engineering student in Electrical and Electronics Engineering at KLS Gogte Institute of Technology. My academic interests lie at the intersection of machine learning (ML) and electronics, where AI-driven technologies can revolutionize embedded systems, automation, power electronics, and robotics. I am particularly fascinated by how ML enhances predictive maintenance in power systems, optimizes energy efficiency, and improves decision-making in control systems. The ability of AI to analyze vast datasets, detect patterns, and enable smart automation aligns perfectly with my passion for innovation in electronics and intelligent systems. I aim to integrate ML into embedded systems to create self-learning devices that adapt to their environments, making industries smarter and more efficient. Beyond academics, I hold leadership positions as the Chair of IEEE PES and Treasurer of IEEE WIE, where I organize research-driven initiatives. I am also an avid basketball player with aspirations of leading my team. My drive to bridge the gap between hardware and AI-powered solutions motivates me to pursue research and hands-on projects that push the boundaries of smart electronics and automation. |
Beta Was this translation helpful? Give feedback.
-
|
@Aryaa-21, I really like how you’ve connected ML with power systems and control. That field has so much untapped potential, and it’s great that you’re already thinking about predictive maintenance and energy efficiency. What caught my eye is your role in IEEE PES and WIE. That’s a fantastic platform to bring some of these ideas to life with others. Even I, as an academic, try to make connections for conferences to embrace new technologies and ideas. Sometimes I feel like we need to be very systematic about it and create growth opportunities; otherwise, conferences can get stuck in the usual routine or topical areas. I’m curious if you’ve thought about how you might use those leadership roles to get more people involved in projects like the ones you care about. |
Beta Was this translation helpful? Give feedback.
-
|
Hello everyone! My name is Renan Michel, and I am a last-year undergraduate student in Control and Automation Engineering at the Federal Institute of Pará (IFPA-Belém). My passion lies in applied robotics, particularly in integrating Machine Learning into robotic systems to enhance autonomy, adaptability, and real-time decision-making. I am deeply fascinated by how intelligent algorithms can empower robots to solve complex real-world challenges, from industrial automation to assistive technologies. What sparked my interest in ML systems:My interest in ML systems was sparked by my fascination with how intelligent algorithms can transform the way machines interact with the world. Machine Learning is revolutionizing countless industries, from healthcare to finance, by enabling data-driven decision-making and automation. ML systems topics that really gets me excited:
The intersection of robotics and ML captivates me. I am particularly interested in developing and optimizing algorithms that enable robots to perceive, learn, and interact more effectively with their environment. Whether through reinforcement learning for robotic control, sensor fusion for real-time decision-making, or TinyML for efficient edge computing in autonomous systems, I am eager to explore ways to push the boundaries of intelligent robotics.
I am also highly interested in evaluating and optimizing ML models for resource-constrained devices. The challenge of balancing accuracy, efficiency, and real-time performance in TinyML models excites me, particularly in robotic applications where computational constraints are a key concern. Developing systematic benchmarking methodologies to assess model performance across different hardware platforms is another area I am keen to explore.
Beyond robotics, I am enthusiastic about deploying AI models on edge devices for autonomous applications. From industrial process automation to smart IoT solutions, I see immense potential in leveraging efficient ML techniques to enable intelligent, self-adaptive systems that operate reliably in dynamic environments. I am looking forward to great discussions and collaborations ahead in the community Find me on LinkedIn: www.linkedin.com/in/renanmichel |
Beta Was this translation helpful? Give feedback.
-
|
I really appreciate how clearly you described your focus on benchmarking for resource-constrained robotics. This type of work is crucial for successful deployments, and it’s great to see someone already considering these aspects. Some time ago, we worked on a benchmark called RobotPerf, similar to MLPerf, which is discussed in https://mlsysbook.ai/contents/core/benchmarking/benchmarking.html. I suggest reviewing the benchmarking chapter in that book. It provides a structured approach to understanding what to measure and why. It could serve as a helpful starting point as you develop your own experiments. I’d also love to hear about the setup or platform you plan to use. |
Beta Was this translation helpful? Give feedback.
-
|
Hello sir, My name is M.Rohit, and I am a research student at the Indian Institute of Technology Madras,working on embedded machine learning. When I first started approaching on TinyML, I was captivated by the idea of running intelligent models directly on microcontrollers—seeing how much could be achieved within such tight memory(constrains) and power budgets really sparked my passion for ML systems.To date, I have implemented ML applications on NVIDIA Jetson Nano, nRF52840, and ESP32 platforms. I have been studying through the MLSysBook and found it valuable—Chapter 2’s deep dive into deployment paradigms (cloud, edge, mobile, tiny) helped me understanding of latency, privacy, and resource constraints more systematically. In particular, I’m keen to explore research questions around hybrid cloud–edge partitioning strategies: for example, how to dynamically decide which parts of a model to run locally versus in the cloud based on real-time network conditions, and usage of ML based upon the device and its constraints. I would be grateful for any guidance on exploring and working around these ideas, especially drawing on the insights . I look forward to learning from all of you and contributing my experiences as I intend to design robust, adaptive deployment pipelines for resource-constrained devices. Warm regards, |
Beta Was this translation helpful? Give feedback.
-
|
@rohitmatta21, the way you’re thinking about hybrid partitioning really gets to the core of what makes ML systems interesting. Figuring out what runs locally and what lives in the cloud isn’t just a technical decision; it’s a systems problem with real consequences for latency, energy, and user experience. Hope the way you decide in Chapter 2 was helpful for that. There was a dedicated section on hybrid models that I added. Also, since you’ve already worked with Jetson, nRF, and ESP32, you might find the kit section of the book helpful. It walks through practical setups for exploring exactly these kinds of tradeoffs. You can find it here: https://www.mlsysbook.ai/contents/labs/kits.html In the future, I'd love to run a lottery or something similar to give people free kits. I used to do that in the past by mailing people kits directly but then i got too busy : ) |
Beta Was this translation helpful? Give feedback.
-
|
Hi everyone, My name is Rafa Aranda, I'm 45 years old and I'm from Spain. My approach now is that of a learner, because I can clearly see the potential of ML in diverse fields and how important it is to find a solid position regarding ML and AI. It's not my first encounter with ML, though. Although my background is on Industrial Engineering, or perhaps because of that, I've been very interested in the possibilities that ML and AI in general can bring to the industrial and the business fronts. I find this ML systems approach to be innovative, and most importantly necessary, since I consider that in order to make a truly creative use of the Al technologies of today one must fully understand not only the conceptual layer but also the hardware and software ones, which are what make up the material substrate upon which ML can be deployed. I'm very excited for having found this valuable "book", which is far more than that, as we are seeing here. Let me thank the author and the many contributors, and express a deep thank you to everyone here for sharing their objectives and vision for learning this material. All the best. |
Beta Was this translation helpful? Give feedback.
-
|
Hi @RafaAA Learning never stops! Good for you that you are still learning. I am the same age as you and every day I make time to learn something new, like working my brain out! If you have the time, be sure to check out https://www.edx.org/professional-certificate/harvardx-tiny-machine-learning, as there are online resources, you don't have to pay for it :) You can audit them for free! |
Beta Was this translation helpful? Give feedback.
-
|
Hello everyone, I have a deep interest in AI for real-world impact, especially in areas like computer vision, robotics, and assistive technologies. I enjoy building intelligent systems that bridge perception and action, combining machine learning with embedded systems and robotics. My work on projects like real-time arrow detection for autonomous navigation (as part of the IRC 2025 team), and the biodegradable material classifier using Raspberry Pi and CNNs, has helped me apply theoretical concepts to physical systems. One of my most impactful experiences was developing an Indian Sign Language Translator during the IBM Z Datathon 2024, which combined NLP, computer vision, and 3D animation to enable communication between hearing and deaf communities. I am also passionate about financial forecasting, demonstrated through my work on a Predictive Market Analysis platform using ML algorithms. I'm always eager to explore new research directions, collaborate with like-minded peers, and contribute to applied AI innovations that address real-world challenges in areas like accessibility, sustainability, and automation. You can find me here: |
Beta Was this translation helpful? Give feedback.
-
|
@Nilangbhuva - that Indian Sign Language Translator project stands out to me. I’d love to hear more about how you approached the pipeline design for it. That kind of insight would be valuable to a lot of people here. |
Beta Was this translation helpful? Give feedback.
-
|
Hi everyone! I’m Shubham Kumar Mandal, a student at IIT Kharagpur, currently pursuing my Dual Degree in Mechanical Engineering. While my core background is in engineering, I’ve always been drawn to the intersection of hardware and intelligent systems—especially how machine learning can be optimized and deployed on real-world devices. My interest in ML systems sparked when I started working on a wearable sound-detection watch for the hearing impaired. I realized that building an ML model is only half the challenge—the real complexity lies in making it run efficiently on a low-power, resource-constrained device. That experience opened my eyes to the world of ML systems. One topic that really excites me is TinyML , I love the idea of pushing intelligence to the edge and making devices smarter without needing cloud access. It’s amazing how much can be achieved with so little hardware, and I’m eager to explore more in this space! Looking forward to learning with all of you and building something impactful together |
Beta Was this translation helpful? Give feedback.
-
|
@shubham21-ai That wearable sound-detection project is a great example of why ML system design matters. Once you step off the whiteboard and try to make a model run on a tiny device, everything becomes very real very fast. I love learning new things from anyone and everyone so I’d love to hear more about how you handled the energy and latency tradeoffs on that watch. |
Beta Was this translation helpful? Give feedback.
-
|
Dear Prof. @profvjreddi and the community, "Intelligence is the log of compute." – Satya Nadella, CEO, Microsoft With a profound passion for exploring the dynamic intersection between computer architecture, hardware-software co-design, chiplets, AI, and reinforcement learning, I am Arnav Shukla Currently pursuing Bachelor’s in Electronics and Communications Engineering at IIIT-Delhi. I aim to achieve the student mindset that is both inventive and deeply analytical. In my research engagements, I am tackling critical AI workload bottlenecks by designing a novel Chiplet-architecture, emphasizing inter-chiplet communication efficiency. I have experience with analog layout optimization, fine-tuning large & RL for post-training language models (LLMs), Conditional GAN-based text-to-image synthesis and a deep reinforcement learning framework for VLSI chip placement and routing through my internship and projects. In my academic journey, I've aimed to move beyond traditional disciplinary boundaries. By engaging with both hardware and software perspectives and staying open to new paradigms and real-world systems, I've tried to build a more holistic understanding of ML systems. I desire innovative mindset, practical experience and never refrain from learning and new things. Looking forward to learn and collaborate with all. |
Beta Was this translation helpful? Give feedback.
-
|
Hi @aethernavshulkraven-allain — I see you’ve left a message here. I didn't realize it as I am just going through quite a few today. Keep up the great work you're doing with the students on the MARL project :-) |
Beta Was this translation helpful? Give feedback.
-
|
Hello everyone! 👋 I'm Durgesh Khade, an third year undergraduate student in IT Engineering at Pune Institute of Computer Technology (PICT), India. My passion lies at the intersection of machine learning systems and real-world engineering — building AI infrastructure that not only works in theory, but scales, adapts, and performs reliably under real-world constraints. Over the past year, I’ve worked on projects that sit at the crossroads of AI, systems engineering, and national-scale problem-solving. As an SDE Intern and Group Leader during a year-long internship with TASL, I contributed to the Indian Navy’s CTS-71 project, translating complex defense requirements into scalable and secure software. My role spanned full-stack development, AI model deployment, and cloud-native systems — building production-grade tools under high-stakes environments. Why ML Systems? I’m deeply curious about what makes models usable and efficient at scale — from deployment pipelines and intelligent scheduling to hardware-aware optimizations and resource-constrained inference. These are the invisible layers that turn ideas into impact, and I’m eager to dive deeper, learn from this community, and contribute meaningfully. Looking forward to connecting, learning, and building together! |
Beta Was this translation helpful? Give feedback.
-
|
@BrewedAlgorithms, that’s a great introduction. You’ve clearly spent time in environments where systems actually have to work, not just run in a lab. Super curious about the CTS-71 project, which must have given you a front-row seat to the kinds of tradeoffs that make ML systems so interesting. What is that project about? |
Beta Was this translation helpful? Give feedback.
-
|
Hi everyone, I’m Varun. I recently finished my Master’s in Computer Science at the University of South Florida, where I worked on machine learning projects in healthcare, including deploying AI models on low-power devices. I got into ML systems when I realized that getting a model to work in the real world is a completely different challenge from training it in a lab. Recently, I’ve been exploring edge AI for cybersecurity, focusing on real-time anomaly detection in safety-critical systems where models need to operate securely and efficiently without cloud dependency. I’m especially interested in edge AI for safety-critical applications, where every millisecond and bit of memory matters. Excited to be here, learn from all of you, and share ideas about making ML models faster, smaller, and more reliable in the real world. I’d love to hear your thoughts or any guidance on exploring this area further. |
Beta Was this translation helpful? Give feedback.
-
|
@VarunSai02, edge AI in safety-critical settings is a space where theory gets tested fast. When the system must react in real time and the cloud isn’t an option, everything from model choice to runtime behavior matters much more. Your focus on anomaly detection fits that world perfectly. You might find it useful to take a look at the book's kits section, since it lays out practical setups for testing those kinds of constraints. https://www.mlsysbook.ai/contents/labs/kits.html. I’d be curious to hear how you’re thinking about security at the edge, especially when it comes to balancing latency with integrity checks. That’s where a lot of the hardest and most interesting problems live. We have actually started doing some research on it. TinyML Security: https://arxiv.org/abs/2411.07114 |
Beta Was this translation helpful? Give feedback.
-
|
Dear Prof. Vijay Janapa Reddi, I have gained a deep understanding of the course content of this project, and it has been of great help to my learning. Therefore, I hope to develop this content into an official Chinese-language book for publication. If this content can indeed be published as a Chinese-language book, I would like to ask for your suggestions or thoughts on this matter. My goal is to make this book a professional resource for chefs in China. Currently, I have a dedicated domestic publishing house in China that I am collaborating with, and I wonder if you have any additional suggestions regarding this publication plan. Best regards, XingChen |
Beta Was this translation helpful? Give feedback.
-
|
I have to say, I did not expect a proposal for translating this book to involve chefs in China. 🙃 That might be the most unexpected deployment use case I’ve heard all year. I suppose model serving could take on a whole new meaning in a kitchen. On a more serious note, I don’t currently have any plans for an official Chinese-language publication with outside partners. The MIT Press is responsible for any book translations. But I’ll definitely remember this one for the story archives. Thanks for giving everyone here a moment of levity. 🤗 |
Beta Was this translation helpful? Give feedback.
-
|
Hi everyone! I love the idea of learning in public with others. I have found that I have an immense amount of curiosity in just connecting with others who have a knwck for poking at things. Seems like I have found a small community here that wmbldies just that. 🙂 I'd love to invite everyone to send a connection request my way. Here is my LinkedIn: http://linkedin.com/in/bcalderonmorales-cmoe And here is my GitHub: https://github.com/BA-CalderonMorales Always happy to collab on ideas. Looking forward to learning with everyone. Thank you Prof. Vijay Janapa Reddi and others for making this open source. 🚀 |
Beta Was this translation helpful? Give feedback.
-
|
Love the energy, @BA-CalderonMorales! Curiosity and a knack for poking at things really capture the spirit of this whole space. It’s what motivates me out of bed at night and keeps me awake too! 🤔 It’s often how the best ideas begin. Glad you found your way here—I have a feeling you’ll fit right in. I truly want to build a community around this. Sometimes I think that maybe dedicating one hour each week to the book could help. That's not to say I have all the answers, but I wonder if that might help us grow more of a community. |
Beta Was this translation helpful? Give feedback.
-
|
Hello everyone! I got into AI and ML as a hobby as I was always into the idea of AI as a kid. This lead to me doing independent research and making online content about what I was working on and getting more passionate about the field. Since I was enjoying learning so much I went back to school leading to where I am now. My schooling is focused more on robotics automation, but my real passion is still AI. This leads to me seeking out ways to learn on my own in addition to school. That's how I found this! I am often frustrated with the limits of my understanding of modern AI. I can train a model and deploy it for specialized tasks, but don't understand the entire process leading to making one and what makes work in its entirety. I'm very pleased to see this resource was created for students with that struggle in mind! I'm glad to have found this resource and would enjoy networking and hope to cross paths with some of you elsewhere in the forms here as we all learn! Thanks for creating this as a resource, I can't wait to dive in. |
Beta Was this translation helpful? Give feedback.
-
|
@vegalyraevt, I love the honesty in how you describe your journey. That frustration you mention is something almost everyone who moves from hobbyist work to deeper systems eventually hits. It’s actually a good sign 💪 It means you’re starting to look under the hood and see how much more there is to understand. A lot of the book was written with exactly that kind of learner in mind (including myself), so I think you’ll get a lot out of it. Every book teaches us something new, one way or the other. I’m looking forward to seeing how your robotics and AI interests come together as you dig in. Sincerely, wishing you the very best! |
Beta Was this translation helpful? Give feedback.
-
|
Hey everyone! I'm Muhammad Hassaan, a passionate master's student in Software Engineering. I’ve gained hands-on experience in the industry, working across automation, IT, data science, and software engineering. To be honest, working in the industry has taught me that building machine learning systems is quite different it’s not just about coding, but about solving problems end-to-end, whatever it takes to ship reliable solutions. I was looking for a resource that explains these concepts clearly and in depth, especially since machine learning connects both hardware and software, making it more complex than traditional software engineering principles or the SDLC cycle. I’m really glad to have found this excellent resource! After reading the preface and a few chapters, I can already tell this book will be my primary reference moving forward. I’m excited to continue learning both as a research enthusiast and a developer |
Beta Was this translation helpful? Give feedback.
-
|
Hello everyone, I am Milind Varma, a software engineer currently working on building computing platforms for ML(SPCS at Snowflake). I have an undergrad degree in Electrical Engineering from IIT Madras and did my Masters in Computer Science from UCSC. As part of my masters I built a RL based cluster level scheduler, capable of dynamic batching and optimized model placement on GPUs, to achieve better GPU resource utilization while satisfying SLOs. I really enjoyed the systems work and want to explore further and deepen my knowledge. Performance engineering really excites me, I like being able to build mental models of whole systems and optimizing for latency and memory. Looking forward to my first foray into TinyML. @profvjreddi Thank you for making this resource open source! :D |
Beta Was this translation helpful? Give feedback.
-
|
Hi All I'm Michel - from Brazil (são paulo, SP). Michel Miotto Barbosa |
Beta Was this translation helpful? Give feedback.
-
|
Hello all! I am Azamat, from Kazakhstan. Now that I see industry growing in Kazakhstan as well, I want to dive in more and hopefully find myself in it. |
Beta Was this translation helpful? Give feedback.
-
|
Hello Everyone, I'm Muhammad Shifa, a Machine Learning Engineer with around 5 years of experience developing AI driven software products related to computer vision and edge AI. Over time, I realized that building accurate models is only part of the journey; making them efficient, scalable, and reliable in production is where true impact lies. My interest in ML systems grew from observing the gap between building accurate models in research environments and deploying them effectively in real-world applications. This realization sparked my interest in ML systems, particularly in understanding the engineering challenges behind deploying and maintaining ML solutions at scale. I’m especially excited about the end-to-end model lifecycle management; including data pipelines to continuous deployment workflows. The challenge of designing systems that ensure both model quality and operational robustness truly motivates me, exploring how we can design ML systems that deliver low-latency, high-efficiency inference on constrained devices while maintaining robustness and scalability across distributed environments. Dear @profvjreddi , Than you for making this one stop solution for my queries. Nicely managed and the most exciting part is the question and answer after each section. This is truly amazing. Thanks to all the contributors. Best regards, |
Beta Was this translation helpful? Give feedback.
and others
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Hi everyone! 👋
I’m Prof. Vijay Janapa Reddi, the author and editor of this resource, and I’m passionate about making foundational technologies like ML/AI accessible to everyone. Along with my students and TAs, I created this resource because we noticed a significant gap: while there are countless books and materials on deep learning theory and concepts, there’s a shortage of resources focused on the ML systems side.
If ML algorithm developers are like astronauts, ML systems engineers are the rocket scientists and mission control specialists who get them to space and keep the mission on track. 🚀 📡
The critical aspects of ML—optimizing models for specific hardware, deploying at scale, and ensuring efficiency and reliability—are often overlooked. These elements transform ML models into real-world applications. Yet, many practitioners struggle due to a lack of available knowledge, not interest.
That’s a little bit about me and my passion.
Now, I’d love to get to know you—our readers and fellow enthusiasts!
Please introduce yourself below and share:
Whether you’re just getting started or have years of experience—your perspective is important to us. Let’s connect, share our insights, and build a vibrant community around machine learning systems together. 🚀
Sincerely,
vj.
Beta Was this translation helpful? Give feedback.
All reactions