# Making AI Smarter by Focusing on Unsure 'Nodes' > This 2025 patent from D5AI LLC describes a way to train AI models more effectively by boosting the learning signal for 'nodes' that aren't making clear decisions on data. - **Patent:** US 12423586 - **Original title:** Training nodes of a neural network to be decisive - **Owner:** D5AI - **Granted:** 2025 - **Status:** Active - **Times cited:** 1 - **Field:** ai_ml, software, semiconductors, consumer_electronics ## What it does This patent explains a method for training artificial intelligence, specifically neural networks. When training an AI, it looks at individual 'nodes' (like tiny decision-makers within the AI) and checks if they are 'decisive' for a given piece of training data. If a node isn't clearly deciding one way or another – meaning its output doesn't strongly lean towards a specific outcome – the system then amplifies the learning signal for that node. This amplification is done by multiplying the 'partial derivative' (a measure of how much the AI's error would change if the node's output changed slightly) by a factor greater than 1.0. This helps the AI learn faster by focusing on the parts that are struggling to make a clear choice, using common training algorithms like stochastic gradient descent. ## What it does NOT cover - Training methods that do not identify if a target node is 'not decisive'. - Methods where the learning signal for an undecided node is not multiplied by a factor greater than 1.0. - Training that only applies to neural networks with fewer than two layers. - Methods that do not involve a 'feed-forward' computation phase. - Methods that do not involve a 'back-propagation' computation phase. - Training where the 'partial derivative' is not computed for the target node. ## The clever bit The innovation lies in identifying AI 'nodes' that are ambivalent or uncertain about their decisions on specific data points, and then specifically boosting their learning rate. Instead of treating all nodes equally during training, it intelligently targets the ones that need more guidance. ## Real-world examples 1. Training large language models 2. Improving image recognition AI 3. Developing AI for autonomous vehicles 4. Optimizing AI in recommendation systems ## Why it matters As AI models become more complex, efficiently training them is crucial. This patent addresses a core challenge in AI development: ensuring that all parts of the model learn effectively. By focusing on nodes that are 'undecided,' this method aims to speed up the learning process and potentially lead to more robust and accurate AI systems, which are increasingly vital across many industries. ## Key terms - **node** — A basic processing unit within a neural network, analogous to a neuron in the brain. - **neural network** — A type of AI inspired by the human brain, made up of interconnected 'nodes' or 'neurons' that process information in layers. - **activation value** — The output value of a node after it processes input data. - **back-propagation** — The process of calculating and assigning error gradients (partial derivatives) back through the neural network layers. - **learned parameters** — The internal settings or weights within a neural network that are adjusted during training to improve performance. - **objective function** — A mathematical function that the neural network tries to minimize during training, representing its error or goal. - **activation function** — A function applied to the output of a node to introduce non-linearity, allowing the network to learn complex patterns. - **partial derivatives** — A mathematical concept used in training AI to measure how a small change in one part of the network affects the overall error. - **optimization algorithm** — A method used to adjust the learned parameters of a neural network to minimize the objective function (e.g., gradient descent). - **feed-forward computation** — The process where input data moves through the neural network layers to produce an output. - **neutral activation value** — A specific output value of an activation function that represents a midpoint or baseline state. - **tanh activation function** — Another common activation function that outputs values between -1 and 1. - **sigmoid activation function** — A common activation function that outputs values between 0 and 1. ## Frequently asked questions ### What does Making AI Smarter by Focusing on Unsure 'Nodes' cover? This 2025 patent from D5AI LLC describes a way to train AI models more effectively by boosting the learning signal for 'nodes' that aren't making clear decisions on data. ### Who owns patent US 12423586? D5AI owns this patent, granted in 2025. ### When does this patent expire? This patent is expected to expire on January 30, 2045, when the invention enters the public domain. ### What is patent US 12423586 cited by? This patent has been cited by 1 later patents that build on its ideas. ### What problem does this patent solve? As AI models become more complex, efficiently training them is crucial. This patent addresses a core challenge in AI development: ensuring that all parts of the model learn effectively. By focusing on nodes that are 'undecided,' this method aims to speed up the learning process and potentially lead to more robust and accurate AI systems, which are increasingly vital across many industries. ### What does this patent NOT cover? Training methods that do not identify if a target node is 'not decisive'. **Full plain-English explainer:** https://patentbrief.org/patent/us/12423586/training-nodes-of-a-neural-network-to-be-decisive **Original patent:** https://patents.google.com/patent/US12423586 --- _Source: PatentBrief — https://patentbrief.org. Patent facts are from public records; the plain-English explanation is PatentBrief's._ ## Related patents Semantically similar inventions in the PatentBrief corpus: - [How to Speed Up Neural Network Training Using Momentum](https://patentbrief.org/patent/us/4914603/training-neural-networks) — A 1988 method for training artificial neural networks that uses an 'activating variable' to speed up how quickly the network learns from its mistakes. - [Training AI Models Across Different Computers](https://patentbrief.org/patent/us/12574477/distributed-deep-learning-using-a-distributed-deep-neural-network) — This 2026 patent describes a way to train AI models on one computer, send a version to another computer for further training with private data, and then update the original model with the improvements. - [Automated AI for Adapting to New Data Without Retraining](https://patentbrief.org/patent/us/20230162023/system-and-method-for-automated-transfer-learning-with-domain-disentanglement) — This patent describes an automated system that builds artificial intelligence models capable of adapting to new, different data without needing full retraining, by learning to ignore irrelevant changes. - [How to Save and Reuse Skills Learned by Artificial Intelligence Hardware](https://patentbrief.org/patent/us/10410117/method-and-a-system-for-creating-dynamic-neural-function-libraries) — A method for capturing the internal settings of a neuromorphic AI chip after it learns a task, allowing that 'skill' to be exported and loaded onto another AI chip. - [How a Camera-Based System Monitors Artificial Neural Network Creativity](https://patentbrief.org/patent/us/10423875/electro-optical-device-and-method-for-identifying-and-inducing-topological-states-formed-among-interconnecting-neural-modules) — A system that uses a camera to watch a screen displaying neural network activity, identifying new patterns and using a critic to decide if those patterns are worth keeping.