Machine Learning for Good: Training Models for Medical Analysis

Leave a reply

The intersection of machine learning (ML) and Healthcare is not just a technological revolution—it’s a profound shift in how we understand and approach human well-being. ML is a tool that’s been evolving quietly behind the scenes for years, but its recent surge in healthcare applications feels like a leap into the future. From diagnostic imaging to personalized treatment plans, we’re witnessing the birth of a healthcare system that’s not only data-driven but capable of adapting to the complexities of the human body and mind.

data in healthcare

At the heart of this transformation is the idea that Healthcare can be predictive rather than reactive. Instead of waiting for symptoms to worsen, we can use machine learning models to analyze subtle cues across multiple forms of data—audio, video, images, and sensor data—to detect conditions like Parkinson’s Disease early on. In a field where time is critical, this capability can be the difference between early intervention and advanced illness.

The Human Factor in ML-Driven Diagnostics

However, it’s easy to get lost in the jargon and overlook the human element behind this revolution. Yes, algorithms can analyze more data in seconds than a doctor might in a lifetime, but these technologies are not about replacing medical professionals—they are about empowering them.

Every pixel, soundwave, and movement analyzed by an ML model carries real human implications. It represents a person’s struggle, their hope for answers, and, ultimately, their health outcomes. By embracing machine learning, we are giving healthcare professionals the tools they need to better understand, diagnose, and treat patients on an individual level.

first car

The healthcare industry’s adoption of ML, particularly in diagnostics, is reshaping the role of doctors from solitary decision-makers to orchestrators of advanced technological tools. The naysayers will paint a picture of a world where machines (or, in today’s terms… AI) will replace humans, but this shouldn’t scare us from embracing these tools. History has shown us that when these new technologies enter our lives, the work doesn’t disappear; it transforms into something new. After all, people said the same thing about automobiles and computers, and look at how that turned out.

Machine Learning Using Multi-Modal Data

What makes this moment even more exciting is the use of multi-modal data—combining information from multiple sources like audio, video, and images. For example, in Parkinson’s Disease diagnosis, an ML model can analyze a patient’s voice, capturing the smallest vocal tremors that may signal early-stage neurodegenerative changes. Simultaneously, video footage of the patient’s movements can be analyzed for physical symptoms, such as tremors or rigidity, that might otherwise go unnoticed in a short clinical visit.

This holistic view of patient data allows for more comprehensive and nuanced diagnoses. It’s not just about analyzing a static image or isolated metric but about building a complete narrative from diverse data sources. These advanced models can sift through the noise and detect meaningful patterns across multiple channels of information, dramatically improving early diagnosis and treatment options.

The Future: Empowering Professionals and Patients

The future of ML in Healthcare isn’t just about technical prowess. It’s about how we as a society choose to harness this power. The goal is not to create a future where machines replace human doctors but one where they augment the capabilities of medical professionals, allowing them to provide more personalized and effective care.

science and technology in medicine

Moreover, these advancements don’t just benefit the healthcare providers. Patients themselves stand to gain significantly, with more accurate diagnoses, earlier interventions, and a more involved role in managing their health. With open access to ML tools and resources, professionals from all backgrounds can build tailored recognition solutions that address their specific needs. The future is about democratizing access to these powerful tools, ensuring more people can benefit from the next wave of medical innovation.

With every new advancement, we’re reminded that this isn’t just about technology—it’s about people. The most exciting part of this journey is how ML is transforming Healthcare not just by numbers and codes, but by improving lives, one model at a time.

—

If you’re interested in learning more about how ML can revolutionize Healthcare, I invite you to attend our Keynote at the 2024 RTC Conference at Illinois Tech, “Training Machine Learning Classification Models for Creating Real-Time Data Points of Medical Conditions,” on Tuesday, October 8, 2024, at 2:45 PM. Dr. Nikki-Rae Alkema, PT, DPT and I will discuss actionable insights into applying ML models in Healthcare with a live demonstration.

2024 RTC Conference at Illinois Tech

You can use the discount code FFSPKR to get $200 off registration. Don’t miss this opportunity to explore the future of machine learning and Healthcare—register today and be part of the conversation!

Top Reasons to Mark Your Calendar for SCaLE Next Year

Leave a reply

In March, I had the fortune of attending and speaking at one of my favorite conferences, Southern California Linux Expo (SCaLE) 21x. As the name suggests, this is the 21st iteration of this tech-heavy yet family-oriented event, which usually takes place in Pasadena but, in some years, in the greater Los Angeles area. This is my sixth time attending (and 3rd time presenting), and I am glad to say that this year’s conference knocked it out of the park again.

What is SCaLE?

SCaLE is North America’s largest community-run open source and free software conference. The entire event, from setting up the networking to managing the session introductions, is all volunteer-based. This allows SCaLE to skip over the pay-for-play sessions you typically see at larger corporate events and focus on quality sessions that attendees are interested in. More importantly, this allows the event to keep the cost of attendance to under $100 for the entire 4-day event and maximum inclusion for those that want to attend.

Southern California Linux Expo 21x

The content ranges from topics like Kubernetes to Open Source AI to the Low-level Linux kernel. My favorite session topics always revolve around IoT/Edge, Security, and anything unique and obscure which you will definitely find a lot of here. I wanted to highlight a few of the more interesting (and hilarious) things I was able to participate in at SCaLE this year. I hope you will enjoy this too…

Kwaai Summit: Personal AI

You want to discuss a very meta but also a very real topic that will arrive at our doorsteps soon: Personal AI. What is Personal AI? It’s the idea that we will have AI systems making decisions on behalf of individuals, or more specifically, you. Whether you know this or not, this is already happening on a small scale (excuse the pun). These are things like your iPhone or Android making reservations at a restaurant or, a more concrete example, making recommendations on things you might be interested in purchasing based on your Instagram or TikTok feed.

Now, imagine we have all this data, information, choices, relationships, and associations to all these different disparate data points. How will these choices and products find their way to grab your attention? In the past decade, it’s been done through associations (when you Instagram heart or Facebook like something) and then extrapolating what else you might enjoy based on probabilities. For example, if you like baseball, you might want to purchase a Dodgers jersey.

Kwaai Summit

The next wave will resemble a personal assistant in the form of an AI agent talking to external AI agents and then making decisions based on those interactions. Your AI agent knows everything about you. What you like, who your friends are, your background, and all other aspects of your life. Based on your unique profile, this AI agent will genuinely know how to interact with your digital environment based on who you are and what apps and access you have.

The Kwaai Summit discussed the new interactions and connections we will have with these AI systems. This was a fascinating series of talks. I recommend checking out The AI Accelerated Personalized Computing Revolution by Pankaj Kedia below.

If we start interacting with the world via proxy using our AI Agents, there will be a lot of interesting fallout from these interactions. First, what controls your AI Agents’ access, and how does it establish trust with these external AI agents? This is important because if these agents act on our behalf, what determines whether these interactions are good and allowed? Second, where did your AI Agent come from? As a precarious scenario, if your agent was created by Amazon, it might steer you to Whole Foods for all your grocery needs. Definite conflicts of interest there.

As a follow-up to this topic, I would check out AI and Trust by Bruce Schneier below. What an interesting future indeed.

Shameless Plug: My Session About Voice AI Assistants

My session at SCaLE was entitled Voice-Activated AI Collaborators: A Hands-On Guide Using LLMs in IoT & Edge Devices. The discussion was framed by landing LLMs and other machine learning models on IoT and Edge Devices and the complications from working in resource-constrained environments, that is, environments with smaller amounts of memory, CPU, etc. When building your IoT or Edge device, you have decisions on how much “work” you want to do on your Edge Device versus remotely in the cloud. More work means more resources. More resources mean a high-priced device.

Since Voice AI Agents, like Alexa, Siri, or Google Home, don’t have traditional graphical user interfaces and solely rely on using spoken word for interaction, the focus of this talk centered around how the transcription accuracy of the commands you give can dramatically impact the quality of the prompt to your LLM or the input to your machine learning models.

If you are interested in learning more about how to optimize running machine learning models at the Edge, check out my recording below:

Turn on the Funnies

I promised something funny, and one of the staples at SCaLE is your annual talk by Corey Quinn. He often pokes fun at topics all throughout the tech industry. He literally does this every single year. It’s tradition at this point. This year’s topic is where I spent a good 7 years of my life dealing with… Kubernetes. A good portion of it is spot on. His talk Terrible Ideas in Kubernetes was another huge success.

SCaLE Recap

Wrapping up an event like SCaLE is no small feat. I would highly recommend attending this conference next year for those who’ve never had the pleasure of attending. What sets SCaLE apart isn’t just its impressive array of sessions ranging from Kubernetes intricacies to the latest in open source AI, but SCaLE stands as a beacon of community, innovation, and inclusivity, and drawing tech enthusiasts from every corner. For me, the biggest draw is to hear from diverse perspectives all throughout the tech industry and meeting new people in a techy social setting.

For those contemplating bringing their families along, you’ll find SCaLE to be an unexpectedly family-friendly event. Imagine sharing your passion for tech while your loved ones enjoy many activities, like Saturday’s Game Night, which offers everything from board games and video games to virtual reality headsets. If you’re based in or near Los Angeles or are looking to attend a conference on the west coast, SCaLE is the place to be with its information-packed sessions, grassroots vibe, and watercooler-style discussions with subject matter experts throughout the industry.

Pushing the Boundaries: My Experience at the Real Time Communication Conference at Illinois Institute of Technology

Leave a reply

This was my first time at the RTC Conference at Illinois Institute of Technology in Chicago, and I was blown away by the caliber of talks at the event. They ranged from very academic and theoretical to extremely practical, capturing what is happening on the ground today in real-time communications. The mix of attendees was one of the most diverse I have ever seen, which included seasoned professionals in their respective fields to bright and inquisitive minds from IIT starting their careers in the workforce. I was fortunate to have some fantastic conversations while I was here, which I will get to at the end of this post, but this was a mind-opening experience that I am grateful to have received.

RTC Conference at Illinois Institute of Technology

My journey to Chicago included presenting two sessions at the conference. The first was titled “Enhancing Real-Time WebRTC Conversation Understanding Using ChatGPT” and the second was “Edge Devices as Interactive Personal Assistants: Unleashing the Power of Generative AI Agents”. The talks each had very different goals in what they were trying to achieve. Based on the feedback and number of questions that I got afterward in the hallway, they were very well received. Attendees got a glimpse into some unique and thought-provoking possibilities they could take home to explore.

Enhancing Real-Time WebRTC Conversation Understanding Using ChatGPT

The upshot of this session was using Generative AI and Large Language Models (LLMs) to influence conversations in real time. The backdrop was using WebRTC as a protocol and platform to host our conversation, but in reality, any medium conducive to carrying a conversation would suffice. However, WebRTC provides an ideal environment as it’s an open standard, and every modern browser has the capability of supporting these voice/video communications.

Large Language Models and Conversation AIs, like ChatGPT, for the first time, have enabled us to influence the conversation on these platforms because they seamlessly participate in conversations and provide relevant contributions to the conversations being had. That’s the key to why this is a “thing” today. These AIs can be proactive in conversation and not just react to them.

If you are interested in learning more and seeing a really cool demo showcasing this in action, look at the recording above. The demo was definitely a crowd-pleaser since we got to highlight two powerful concepts: AI can retain the history of the conversation taking place and meaningfully participate to influence the simulated conversation for the demo.

All of these resources, links to articles mentioned, and open source projects used in this presentation can be found in the slides. There are also instructions on how to reproduce the demo within this talk.

Edge Devices as Interactive Personal Assistants: Unleashing the Power of Generative AI Agents

My second session focused on Autonomous AI Agents. This might be an unfamiliar topic to some, but we have all heard about them interacting in the real world. Unlike Siri and Alexa, which focus on a single transactional question and response, these are processes where AI models can create their own sub-tasks for problems that need more refinement or detail that a single answer might not be able to answer sufficiently. Without these Autonomous Agents, we typically achieve this refinement by asking the AI ourselves to drill down into a problem further. In this case, the autonomous agent process provides its own questions to seek out the details of the answer.

Since these LLMs are getting smarter and consume fewer resources than previous generations, they can live on IoT (Internet of Things) and Edge devices for the first time thanks to devices getting denser with more hardware capabilities and resources. This talk focuses on different architectures that can be used to land these Autonomous Agents on these IoT/Edge devices to focus on jobs that could run for many minutes to multiple days. The trade-off in these cases is answers with concentrated amounts of knowledge and a more thorough response versus the speed of the reply.

Take a look at the recording of the session above. There is also a demo at the end of this presentation that serves as a proof-of-concept to demonstrate what could be done with these Autonomous AI Agents using an open source project I wrote called Open Virtual Assistant. The demo highlights exercising the “memory” for these Agents (via their vector database) and how one might launch these agents within an IoT or Edge device.

Again all of these resources, links to articles mentioned, and open source projects used in this presentation can be found in the slides with instructions on reproducing the demo within this talk.

Looking Back… Personal Reflections

My most memorable parts of the conference were the conversations with other attendees. The last time I attended a conference was in the second half of 2022, just before ChatGPT blew up in the media. Oh, how times have changed! Besides a good number of the talks focusing on ChatGPT or AI in general, the buzz was definitely in the air and dominated most of the chatter in the hallways. I love tech talk, but I might love these philosophical, what-if, and future-predicting conversations even more.

One of the top questions I get asked frequently as someone who works in the AI/ML is, “Will humans lose our jobs to AI?”. I was also asked this while at the conference. My response is always yes and no. Yes, certain jobs will likely become obsolete… and No, there will always be jobs out there since we haven’t solved all the problems contained within our existence. There will always be jobs, but they might look completely different from what they are today.

An example I always like to give is the calculator and personal computer. When the calculator became mainstream, did all accounting or anything related to math disappear? The answer is definitely No. Society then created far more advanced “calculators” in the form of personal computers that do a bunch of repetitive tasks and naturally transitioned into automation. This is the automation of building cars, canning foods, or mass-producing clothing or shoes. This automation obsoleted people canning foods but created new jobs to develop and maintain these new automation systems.

Everyone recognizes the potential for Artificial Intelligence as the next significant disruptor to humanity. My advice is that just like the calculator or computer, it’s best to understand and know how to use these systems regardless of your field. Those who know how to leverage AI will outshine those who don’t. Finally, when there isn’t a need for a door-to-door encyclopedia salesperson, it’s best to objectively recognize the change in tides and learn something new to transition to. That’s my long-winded answer.

Meeting New Friends

I had a great time at RTC Conference at IIT, and if you are a person like me who loves to learn new and exciting topics, technologies, and ideas, then this is a great place to expand your mind. I highly recommend going and hope I am lucky enough to be working on something compelling enough to share with others next year. Cheers!