Registre des initiatives en IA

30 initiative(s) courante(s)

• 226

  DFO AI - Pipeline

  DFO Pipeline Analyzer is a private Azure DevOps extension designed to assist development teams in diagnosing and resolving pipeline failures using AI. When a pipeline run fails, this task automatically triggers, extracts the provided error details, and queries a secure AI API to generate a human-readable explanation of the failure along with actionable troubleshooting steps. It will assist developers to make appropriate steps in right direction as they are getting suggestion with-in pipeline logs. Suggestion also provide corrected version of pipeline Key features: Runs only on pipeline failure (condition: failed()) Fully agent-agnostic: Works on self-hosted and Azure-hosted agents, across Windows, Linux, and macOS Provides clear error analysis and resolution guidance directly in pipeline logs Keeps your data secure by using your own OpenAI or other AI API key Fully private and scoped to your Azure DevOps organization We have working prototype of DFO-AI pipeline Analyzer by communicating with OpenAI securely. In next steps we are also exploring to integrate with our self-hosted AI Models (open-source via Ollama), currently in discussion

  Pilote Dirigeant principal du numérique Infrastructure et habilitation

• 113

  Improving forecasting for high-priority ports

  Research into feasibility of improving accuracy and timeliness of short-term forecasts for coastal regions by combining traditional and AI modelling methods.

  Pilote Sciences des écosystèmes et des océans Automatisation et aide à la décision

• 99

  Improved freshwater runoff modelling in Eastern Canada to drive ocean models

  Use Neural Networks to post-process streamflow simulations from the WRF-Hydro model, improving their agreement with observational data. These neural networks are applied as a downstream calibration step. Builds synergy between traditional calibration methods and modern AI techniques in hydrologic modeling.

  Pilote Sciences des écosystèmes et des océans Automatisation et aide à la décision

• 102

  Past, Present and Future Oceanographic Conditions in Canada's MPAs and Marine OECMs (Other Effective area-based Conservation Measures)

  Downscaling climate models using neural networks for conservation areas. Neural networks allow us to represent coarse resolution climate projections at a much higher spatial resolution across six eastern MPAs.

  Pilote Sciences des écosystèmes et des océans IA cognitive et générative

• 110

  Mercator Collaboration on Digital Twins of the Ocean

  Knowledge exchanges with ECCC and Mercator Océan International on use of AI for ocean modelling, in the context of Digital Twins of the Ocean.

  Pilote Sciences des écosystèmes et des océans Infrastructure et habilitation

• 111

  Digital Twins of the Ocean Task Team

  National exchanges on topics relating to ocean modelling for development of Digital Twins of the Ocean, including use of AI for modelling and products.

  Pilote Sciences des écosystèmes et des océans Infrastructure et habilitation

• 112

  ECCC exchanges on use of AI for ocean modelling

  Exchanges with ECCC on implementation of machine learning methods for operational ocean forecasting, including potential uses for DFO models.

  Pilote Sciences des écosystèmes et des océans Infrastructure et habilitation

• 132

  Salmon Scale Aging​

  A model using computer vision to look at images of salmon scale and predict the age of salmon and also explain how it arrived at its prediction.

  Pilote Sciences des écosystèmes et des océans Perception et compréhension

• 135

  PII & Sensitivity Scores​

  Determines the sensitivity of the content of a document based on a trained model and allows for the redaction of the sensitive material in the document.

  Pilote Sciences des écosystèmes et des océans Automatisation et aide à la décision

• 137

  Method Recommendation Engine​

  A large language model reviews a scientist’s objective for collecting data in the field and based on the criteria, the model will provide the best procedure to use for their work

  Pilote Sciences des écosystèmes et des océans Automatisation et aide à la décision

• 144

  Factoid Finder Tool

  This completed project used the Factoid Finder extracts text from machine-readable PDFs to create searchable content libraries. Small Language Models are used to implement semantic search techniques, allowing users to search the library for relevant passages within the PDFs.

  Pilote Sciences des écosystèmes et des océans IA cognitive et générative

• 152

  Data Validation Tool

  Use large language models to help detect errors in data transcribed using optical character recognition. Apply validation rules to propose corrections and normalize data into a standardized format.

  Pilote Sciences des écosystèmes et des océans Automatisation et aide à la décision

• 161

  Remote sensing of the internal tide in the St. Lawrence

  Using model outputs (STLE500/STLE200) to 1) help identify internal tide surface signature in SWOT (Surface Water and Ocean Topography) data and 2) train an AI to detect internal tide in SWOT data.

  Pilote Sciences des écosystèmes et des océans Perception et compréhension

• 170

  NFIS Elver Retrieval-Augmented Generation Pilot with CDOS

  CDOS RAG pilot accept NFIS submission of illegal elver sales proposal. AI model would receive content and asked to summarize or visualize information based on core concepts to draw inference. Pilot only active for 2 months. Governance: Large project of 40 pilots led by CDOS. Assign project lead; document user access. Regular check-ins with IT and management. Guardrails: Restrict to pilot users; apply GC data/privacy standards; conduct ethical reviews and audits. No interne

  Pilote Secteur des programmes IA cognitive et générative

• 51

  EN/FR Translation Model​

  A large language model that is trained on the Translations Bureau’s and DFO's language-characteristics (i.e. grammar, annotation, syntax, words, scientific term) context. The model’s enables translation to be like those by the Translation Bureau. ​​Training on large database of scientific documents (published by Canadian Science Advisory Secretariat) to understand scientific terminology is currently in progress. MADLAD 400 10B (Initially trained on sample of ~100 translated documents, in progress of additional training with published CSAS documents going back to year 2000)

  Pilote Secteur des programmes Interaction et engagement

• 78

  Okanagan sockeye life cycle model

  Using LLM (ChatGPT) to support the development of a statistical lifecycle model for Okanagan sockeye. LLMs are not developing the models but are used to brainstorm modelling approaches, debug code, and confirm that written model descriptions in manuscripts are accurate and clear. AI is not being tasked with developing these models. Large Language Models

  Pilote Sciences des écosystèmes et des océans IA cognitive et générative

• 63

  Factoid Finder Tool

  The Factoid Finder extracts text from machine-readable PDFs to create searchable content libraries. Small Language Models are used to implement semantic search techniques, allowing users to search the library for relevant passages within the PDFs. Small Language Model and Text Embedder: MS Marco Distilbert Dot-v5 Cross-encoder: MS Marco MiniLM-L6-v2

  Pilote IA cognitive et générative

• 65

  Comprehensive marine substrate classification applied to Canada’s Pacific shelf

  We built five regional and one coastwide substrate model for the BC coast using substrate observations and seafloor bathymetry derivatives and oceanographic predictors using Random Forests. Random Forests

  Pilote Sciences des écosystèmes et des océans Automatisation et aide à la décision

• 32

  CTD Anomaly Detection

  The goal of this project is to apply machine learning to detect anomalous oceanographic data, highlighting meaningful physical phenomena in the state of the ocean for closer investigation by scientists. A proof of conept model has been developed to demonstrate feasibility. Further validation is required with the client to verify alignment with the needs of scientists. Client: Pacific Region - Ocean Sciences Division

  Pilote Politiques stratégiques Perception et compréhension

• 134

  LLM Document Chatbot​

  Large language model that can help users ask questions to document sets and get answers, summarize key information, and find data with citations and sourcing. ​

  Développement Sciences des écosystèmes et des océans Interaction et engagement

• 53

  LLM Document Chatbot​

  Large language model that can help users ask questions to document sets and get answers, summarize key information, and find data with citations and sourcing., including, from documents with multiple templates using prompt engineering approaches Azure OpenAI GPT4o/4o mini

  Développement Secteur des programmes Interaction et engagement

• 175

  AI model for duplcate or dubious licence accounts

  Licencing systems hold 10 times the amount accounts compared to licences issued. Many duplicate and dubious accounts across the millions of accounts. An AI model to discover duplicate/dubious accounts to highlight them for human review and action.

  Idéation Secteur des programmes Automatisation et aide à la décision

• 172

  AI model for Condtions of Licences review

  Pac has 130 COls. These have been individually edited over the past decades. An AI model to search and identify difference in wording for similar Conditions

  Idéation Secteur des programmes Automatisation et aide à la décision

• 71

  News topic detection tool

  Use large language models and topic clustering to read topics emerging in the news and detect which of those topics is relevant to DFO, as well as what knowledge DFO has available to address those topics. Large Language Models

  Idéation Sciences des écosystèmes et des océans IA cognitive et générative

• 75

  Study on the use of Artificial Neural Networks to Predict the Return Timing and Northern Diversion Rate for Migrating Fraser River Sockeye Salmon

  The Machine Learning feasibility study compliments the statistical approach by asking whether Artificial Intelligence (AI) methods can be developed to predict salmon behaviour as functions of ocean conditions. classical machine learning models, including linear regression, Ridge regression, and Random Forest

  Idéation Sciences des écosystèmes et des océans Automatisation et aide à la décision