{
  "patent_number": "US 12327190",
  "country": "US",
  "title": "AI System for Analyzing Mixed Financial Information",
  "original_title": "Multimodal financial technology deep learning core with joint optimization of vector-quantized variational autoencoder and neural upsampler",
  "summary": "This patent describes an AI system that combines different types of financial data, like stock prices, news articles, and market sentiment, to make better predictions and create new financial scenarios.",
  "what_it_does": "This system processes various kinds of financial information, such as stock prices (time-series data), news articles (textual data), market mood (sentiment data), and company reports (structured tabular data) (Claim 2). It uses specialized AI tools, called 'modality-specific codeword generators' (Claim 1), to convert each data type into a unique digital code. These codes are then blended together by a fusion module to understand their relationships (Claim 1). A 'VQ-VAE encoder' (Claim 1) compresses this combined information into a smaller, discrete code, which is then rebuilt by a 'VQ-VAE decoder' (Claim 1). A 'neural upsampler' (Claim 1) then enhances the reconstructed data, adding back details that might have been lost during compression. All these parts are 'jointly trained' (Claim 1) simultaneously using a 'joint loss function' (Claim 1) to ensure they work together perfectly. For example, a financial firm could use this to predict stock movements by combining historical prices, company news, and social media sentiment, and then generate new, realistic crisis scenarios to test investment strategies.",
  "what_it_does_not_cover": [
    "Does not cover systems that process only one type of financial data, as it specifically requires 'multimodal data' (Claim 1).",
    "Does not cover systems that do not use a 'Vector-quantized Variational Autoencoder (VQ-VAE)' for data compression (Claim 1).",
    "Does not cover systems that lack a 'neural upsampler' specifically designed to recover information lost during compression (Claim 1).",
    "Does not cover AI models where the VQ-VAE and neural upsampler are trained separately, as it requires 'jointly train[ing]' them with a combined loss function (Claim 1).",
    "Does not cover general multimodal AI systems that are not specifically applied to 'financial data' (Claim 2).",
    "Does not cover systems that do not encode each data modality into a 'codeword-based representation' (Claim 1)."
  ],
  "filed": "2024-09-17",
  "granted": "2025-06-10",
  "expires": "2044-09-17",
  "status": "active",
  "holder": "Atombeam Technologies",
  "holder_url": "https://patentbrief.org/company/atombeam-technologies",
  "inventors": [
    {
      "name": "Brian Galvin",
      "url": "https://patentbrief.org/inventor/brian-galvin"
    }
  ],
  "times_cited": 14,
  "tags": [
    "finance",
    "ai_ml",
    "software",
    "telecommunications"
  ],
  "abstract": "A system and methods for processing diverse financial data types using a jointly trained Vector-quantized Variational Autoencoder (VQ-VAE) and neural upsampler. This system efficiently handles time-series, textual, sentiment, and structured tabular data through specialized encoding modules. A novel fusion module integrates these encodings, capturing cross-modal relationships via attention mechanisms and gated fusion units. The fused representation is compressed into a discrete latent space by the VQ-VAE encoder, then reconstructed and enhanced by the VQ-VAE decoder and neural upsampler, respectively. Joint training optimizes all components simultaneously, using a comprehensive loss function that balances reconstruction quality across modalities with upsampling performance. This approach enables superior data compression, reconstruction, and analysis, leveraging inter-modal correlations to improve financial forecasting, risk assessment, and decision-making. The system's ability to explore the latent space facilitates generation of new, synthetic financial scenarios for robust model testing and strategy development.",
  "url": "https://patentbrief.org/patent/us/12327190/multimodal-financial-technology-deep-learning-core-with-joint-optimization-of-ve",
  "markdown_url": "https://patentbrief.org/patent/us/12327190/multimodal-financial-technology-deep-learning-core-with-joint-optimization-of-ve/md",
  "google_patents_url": "https://patents.google.com/patent/US12327190",
  "relatedPatents": [
    {
      "patentNumber": "20250307927",
      "countryCode": "US",
      "title": "How AI Analyzes Financial Data Across Different Timeframes",
      "url": "https://patentbrief.org/patent/us/20250307927/multi-scale-temporal-attention-processing-system-for-multimodal-deep-learning-wi"
    },
    {
      "patentNumber": "12307349",
      "countryCode": "US",
      "title": "How an AI System Answers Financial Questions Using Specialized Bots",
      "url": "https://patentbrief.org/patent/us/12307349/systems-and-methods-of-large-language-model-driven-orchestration-of-task-specifi"
    },
    {
      "patentNumber": "11922469",
      "countryCode": "US",
      "title": "How Computers Rank Financial News for Analysts",
      "url": "https://patentbrief.org/patent/us/11922469/automated-news-ranking-and-recommendation-system"
    },
    {
      "patentNumber": "11615208",
      "countryCode": "US",
      "title": "How Cloud Systems Automatically Create and Train AI Data Models",
      "url": "https://patentbrief.org/patent/us/11615208/dall-e-text-to-image-generation"
    },
    {
      "patentNumber": "11170293",
      "countryCode": "US",
      "title": "How AI Systems Learn to Predict and Act Simultaneously",
      "url": "https://patentbrief.org/patent/us/11170293/alphago-policy-and-value-networks"
    }
  ]
}