Low-Bandwidth Self-Improving Transmission of Rare Training Data

A severe bandwidth mismatch between incoming sensor data rate and wireless backhaul bandwidth often exists on unmanned probes when collecting new training data for machine learning (ML). To overcome this mismatch, we describe a self-improving ML-based transmission system called Hawk. Starting from a weak model that is trained on just a few examples, it seamlessly pipelines semi-supervised learning, active learning, and transfer learning, with asynchronous bandwidth-sensitive data transmission to a distant human for labeling. When a significant number of true positives (TPs) have been labeled, Hawk trains an improved model to replace the old model. This iterative workflow, called Live Learning, continues until a sufficient number of TPs have been collected. For very rare events on challenging datasets, and bandwidths as low as 12 kbps, a team of 7 probes using Hawk discovers up to 87% of the TPs that could have been discovered via full preview, transmission and labeling of all mission data. Hawk also uses diversity sampling and few-shot learning.

• Publication year

  2023

• Venue

  ACM/IEEE International Conference on Mobile Computing and Networking

• Publication date

  2023-10-02

• Fields of study

  Computer Science, Engineering

• Identifiers
  DOI 10.1145/3570361.3613300
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• Edge-based Privacy-Sensitive Live Learning for Discovery of Training Data

  2023cited by this paper
• Programmable microfluidics for dynamic multiband camouflage

  2023cited by this paper
• Low-Bandwidth Remote Sensing of Rare Events

  2023cited by this paper
• Variable Few Shot Class Incremental and Open World Learning

  2022cited by this paper
• Learning Rare Category Classifiers on a Tight Labeling Budget

  2021cited by this paper
• Few-shot Open-set Recognition by Transformation Consistency

  2021cited by this paper
• Few-Shot Incremental Learning with Continually Evolved Classifiers

  2021cited by this paper
• Few-Shot Learning With Class Imbalance

  2021cited by this paper
• Deep Long-Tailed Learning: A Survey

  2021cited by this paper
• Few-shot imbalanced classification based on data augmentation

  2021cited by this paper
• Background Splitting: Finding Rare Classes in a Sea of Background

  2020cited by this paper
• Overcoming Classifier Imbalance for Long-Tail Object Detection With Balanced Group Softmax

  2020cited by this paper
• Curriculum Labeling: Revisiting Pseudo-Labeling for Semi-Supervised Learning

  2020cited by this paper
• Distribution Aligning Refinery of Pseudo-label for Imbalanced Semi-supervised Learning

  2020cited by this paper
• Class-Balanced Loss Based on Effective Number of Samples

  2019cited by this paper
• Underwater Sensor Networks : Applications , Advances and Challenges

  2019cited by this paper
• Diverse mini-batch Active Learning

  2019cited by this paper
• The Computing Landscape of the 21st Century

  2019cited by this paper
• Large-Scale Long-Tailed Recognition in an Open World

  2019cited by this paper
• A comparative study of LPWAN technologies for large-scale IoT deployment

  2019cited by this paper
• Detection of Marine Animals in a New Underwater Dataset with Varying Visibility

  2019cited by this paper
• Pseudo-Labeling and Confirmation Bias in Deep Semi-Supervised Learning

  2019cited by this paper
• Federated Learning: Challenges, Methods, and Future Directions

  2019cited by this paper
• Learning to Sample: An Active Learning Framework

  2019cited by this paper
• Exploring the Limits of Weakly Supervised Pretraining

  2018cited by this paper
• DeepIoT: Compressing Deep Neural Network Structures for Sensing Systems with a Compressor-Critic Framework

  2017cited by this paper
• Lifelong machine learning: a paradigm for continuous learning

  2017cited by this paper
• Focal Loss for Dense Object Detection

  2017cited by this paper
• DOTA: A Large-Scale Dataset for Object Detection in Aerial Images

  2017cited by this paper
• Mars orbital image (HiRISE) labeled data set

  2017cited by this paper
• Learning Cumulatively to Become More Knowledgeable

  2016cited by this paper
• Identifying and Mitigating Labelling Errors in Active Learning

  2015influential reference
• Deep Residual Learning for Image Recognition

  2015cited by this paper
• Never-Ending Learning

  2015cited by this paper
• Capturing Long-Tail Distributions of Object Subcategories

  2014cited by this paper
• “Utilizing” Signal Detection Theory

  2014cited by this paper
• Pseudo-Label : The Simple and Efficient Semi-Supervised Learning Method for Deep Neural Networks

  2013cited by this paper
• Active learning

  2013cited by this paper
• Recognizing linked events: Searching the space of feasible explanations

  2009cited by this paper
• Curriculum learning

  2009cited by this paper
• Density-Based Clustering over an Evolving Data Stream with Noise

  2006cited by this paper
• Lifelong robot learning

  1993cited by this paper

• Edge-Based Live Learning for Robot Survival

  2025influential citation
• Using Radar for Edge-based Live Learning

  2025influential citation
• Pairwise Maximum Likelihood For Multi-Class Logistic Regression Model With Multiple Rare Classes

  2025cites this paper
• FrameScope: Temporal Data Valuation for Stream Active Learning in Autonomous Vehicle Systems

  2025cites this paper
• Link-Aware Energy-Frugal Continual Learning for Fault Detection in IoT Networks

  2025influential citation
• Beyond Federated Learning: Survival-Critical Machine Learning

  2024cites this paper