{
  "patent_number": "US 4914603",
  "country": "US",
  "title": "How to Speed Up Neural Network Training Using Momentum",
  "original_title": "Training neural networks",
  "summary": "A 1988 method for training artificial neural networks that uses an 'activating variable' to speed up how quickly the network learns from its mistakes.",
  "what_it_does": "The patent describes a way to train neural networks by splitting each neural unit into two parts. The first part handles the primary learning, while the second part tracks changes over time to create an 'activating variable.' This variable acts like momentum; it is added to the feedback signal to accelerate how quickly the network adjusts its internal variables (weights) to reach a correct solution. By comparing the network's output to a desired target and iterating through examples, the system converges on a result faster than standard methods of the time.",
  "what_it_does_not_cover": [
    "Does not cover modern deep learning architectures like Transformers or Convolutional Neural Networks.",
    "Does not cover hardware-specific implementations like GPUs or TPUs.",
    "Does not cover unsupervised learning techniques where no desired output is provided.",
    "Does not cover backpropagation algorithms that do not utilize this specific two-subunit momentum mechanism."
  ],
  "filed": "1988-12-14",
  "granted": "1990-04-03",
  "expires": "2008-12-14",
  "status": "expired",
  "holder": "GTE Laboratories Inc",
  "holder_url": "https://patentbrief.org/company/gte-laboratories-inc",
  "inventors": [
    {
      "name": "Laurence F. Wood",
      "url": "https://patentbrief.org/inventor/laurence-f-wood"
    }
  ],
  "times_cited": 29,
  "tags": [
    "ai_ml",
    "software",
    "semiconductors"
  ],
  "abstract": "A method of training an artificial neural network uses a computer configured as a plurality of interconnected neural units arranged in a layered network including an input layer having a network input, and an output layer having a network output. A neural unit has a first subunit and a second subunit. The first subunit having one or more first inputs, and a corresponding first set of variables for operating upon the first inputs to provide a first output. The first set of variables can change in response to feedback representing differences between desired network outputs for selected network inputs and actual network outputs. The second subunit has a plurality of second inputs, and a corresponding second set of variables for operating upon said second inputs to provide a second output. The second set of variables can change in response to differences between desired network outputs for selected network inputs and actual network outputs. The computer provides an activating variable representing the difference between current second output and previous second outputs. A series of examples of data is provided as network input to said network. The activating variable is added to the feedback to accelerate the change of said first set of variables. The actual resulting network outputs are compared to desired outputs corresponding to the examples. The examples are iterated until the network outputs converge to a solution.",
  "url": "https://patentbrief.org/patent/us/4914603/training-neural-networks",
  "markdown_url": "https://patentbrief.org/patent/us/4914603/training-neural-networks/md",
  "google_patents_url": "https://patents.google.com/patent/US4914603",
  "relatedPatents": [
    {
      "patentNumber": "12423586",
      "countryCode": "US",
      "title": "Making AI Smarter by Focusing on Unsure 'Nodes'",
      "url": "https://patentbrief.org/patent/us/12423586/training-nodes-of-a-neural-network-to-be-decisive"
    },
    {
      "patentNumber": "11741188",
      "countryCode": "US",
      "title": "How a Chip Uses Memory to Speed Up AI Calculations",
      "url": "https://patentbrief.org/patent/us/11741188/hardware-accelerated-discretized-neural-network"
    },
    {
      "patentNumber": "10956815",
      "countryCode": "US",
      "title": "How to Fix Faulty Memory Cells in AI Chips",
      "url": "https://patentbrief.org/patent/us/10956815/killing-asymmetric-resistive-processing-units-for-neural-network-training"
    },
    {
      "patentNumber": "10832138",
      "countryCode": "US",
      "title": "How to Automatically Expand Neural Networks by Adding New Nodes",
      "url": "https://patentbrief.org/patent/us/10832138/gpt-language-model-pre-training"
    },
    {
      "patentNumber": "11544573",
      "countryCode": "US",
      "title": "How Projection Neural Networks Speed Up AI Predictions",
      "url": "https://patentbrief.org/patent/us/11544573/llama-large-language-model-architecture"
    }
  ]
}