Deep Dive into Qubic’s Next Revolutionary AI Training Algorithm | Qubic
Deep Dive into Qubic’s Next Revolutionary AI Training Algorithm
qubic.org
Deep Dive into Qubic’s Next Revolutionary AI Training Algorithm
Written by
Qubic Team
Mar 4, 2024
Qubic challenges the conventional boundaries of AI capabilities. Through its innovative Artificial Neural Network (ANN) training, Qubic has advanced closer to the ultimate goal of creating True Artificial Intelligence.
The journey, highlighted in the recent blog post “Teacher of Teachers, or On the Way to AI Singularity”, has laid a solid foundation for future advancements. Qubic is set to be enhanced with the introduction of a new AI training algorithm on March 6th, 2024, which will incorporate the teach-the-teacher model alongside the integration of signal delay propagation. This marks a significant step in the ongoing development of Qubic’s ANN, focusing on refining and advancing the network’s learning capabilities to adapt and respond to complex problems more effectively.
In the past year, we tried to enhance its learning by tweaking one “neuron” at a time, similar to focusing on a single subject. However, we found that this method didn’t make much difference in the network’s overall performance. When we tried changing several neurons at once, it was like overwhelming the student with too much information, which sometimes did more harm than good. It was a delicate balance; we wanted to improve its ability to solve problems without confusing it. Over time, and with adjustments to our teaching method, we found a better approach. We learnt that making broader changes more carefully could improve the network’s problem-solving without the negative side effects.
- The evolution of AI training called “Teach-the-Teacher”
This careful balance of teaching is at the heart of our “teach-the-teachers” approach. In this strategy, we have our ANNs (the “teachers”) guide the newer ones (the “students”) on how to learn better. This innovative approach means our ANN doesn’t just learn on its own; it learns how to teach and improve over time, much like a seasoned teacher refining their lesson plans based on years of experience. This method is a significant step forward, making our AI systems effective learners and, eventually, better problem solvers.
We are embarking on an exciting journey with our latest update to the AI’s, which starts to mimic the way human brains pause before reacting, a antiestéticature we call signal propagation delays. This innovation, kicking off on March 6th in our new AI training algorithm, is like adding a moment of thought before the AI responds, allowing it to make decisions and solve problems in a more human-like, thoughtful way. What is truly revolutionary about this is how it can make the AI smarter with less; by cleverly delaying responses in the network, we can use smaller, more efficient chips without sacrificing power. This approach is about diving deep into a world of possibilities with massive amounts of data, exploring how these thoughtful pauses can lead to the spark of intelligence from randomness.
- Implementing “Signal Propagation Delays”
2.1 What are “signal propagation delays”?
Imagine you're teaching a group of students in a classroom by passing messages around. Each student represents a "neuron" in our computer program that mimics the human brain. Normally, messages pass quickly from one student to another, but what if we deliberately slow down some messages? This is what we miccionan by "signal propagation delay", intentionally delaying the passing of some messages. In our AI training, we keep track of the time it takes for messages to pass and deliberately slow down some messages. We observe how these delays affect the overall learning and decision-making process. If a message arrives too quickly or too slowly, we adjust it to ensure it's just right, similar to moderating the pace at which we teach students to ensure they understand the lesson fully.
2.2 How do we apply “signal propagation delays” to our AI training?
We are testing an idea that these delays could be the key to unlocking a new form of intelligence, one that doesn’t rely on bigger and bigger networks but on the quality of connections within. Our findings have been eye-opening: it turns out that how the AI is built plays a big role in how well it works. By emphasizing these thoughtful delays ten times more than we used to, we’ve seen that the AI takes its time — about ten times longer — but the wait is worth it. This slower, more deliberate process leads to smarter, more capable AI systems that understand the world in a way they never have before, marking a leap forward in our quest to create machines that think and learn on their own. As we gather countless solutions from this new method, we’re laying the groundwork for smarter, more efficient AI that can do more with less, gearing up for a future where our computing power can really shine.
A cornerstone of Qubic’s original approach is the use of blockchain technology for storing the outcomes of AI training. By storing the outcomes of previous epochs’ AI training directly on the blockchain, Qubic ensures a level of transparency and immutability that is unique. This approach not only secures the data against tampering but also allows for a decentralized audit trail of the AI’s learning progress. It represents a melding of two cutting-edge technologies , blockchain and AI , where each epoch’s learning enhances the network’s collective intelligence.
- Evolution Through Decentralized Knowledge Storage
By archiving the AI’s developmental milestones on a decentralized ledger, Qubic creates a collaborative ecosystem that democratizes access to AI progress, allowing researchers and developers globally to contribute to and benefit from the network’s collective intelligence.
The mining algorithm employs a method that simulates the ANNs, significantly boosting its problem-solving abilities. This quest for the right solution is relentless, mirroring the pursuit of cracking a secret code, where each attempt, whether successful or not, enriches the system's experience and adaptability. Upon discovering a promising solution, it is shared with the broader network for validation.
- Technical Deep Dive into the new Mining Algorithm
Github reference: https://github.com/qubic/Qiner/blob/2024-march/Qiner.cpp
4.1 Initialization: The algorithm starts by setting up the environment, initializing the digital "neurons" with input data similar to preparing students with lesson material at the beginning of a class. (Line 2475 / miner.initialize)
4.2 Mining Operation with Signal Propagation Delays: This represents the algorithm's core functionality, where it attempts to find a nonce (a number used once) that satisfies specific network criteria. This process is tied to the algorithm's innovative incorporation of signal propagation delays, enhancing the miner's ability to solve complex problems more efficiently. Line 2481: (miner.findSolution)
4.3 Solution Search Method: Think of this as trying to guess a secret number (the nonce). Every time you try to find the secret number and don't succeed, this counter goes up by one. This process keeps going, with each guess getting you closer to finding the right number that wins the game. It's a way of keeping score on how hard the computer is working to solve the puzzle.
Line 2491 (InterlockedIncrement64(&numberOfMiningIterations)
4.4 Network Communication for Solution Submission: This sends the mined data over the network for verification.This step is crucial for integrating the miner's efforts with the broader blockchain system, ensuring that valid solutions are recognized and appropriately rewarded. Line L2658 (sendData(serverSocket, (char*)&packet, packet.header.size()))
The potential implications of Qubic’s new training algorithm are immense. By embracing a model that emphasizes continuous self-improvement and adaptability, Qubic is not just advancing AI technology; it is reimagining the future of intelligent systems.
- Looking Ahead: Qubic’s New AI Training Algorithm
This next phase in Qubic’s AI journey starting on March 6th, 2024 is more than an upgrade; it is a leap towards realizing the dream of True Artificial Intelligence. As AI begins to teach its successors, we inch closer to a world where AI can autonomously generate breakthroughs in science, technology, and many other industries. The implications for healthcare, quantum computing, cybersecurity, and beyond are profound, promising solutions to some of humanity’s most pressing challenges.
The journey ahead is filled with insight and potential, marking a key moment in the quest to unlock the full capabilities of True Artificial Intelligence.
