Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ScarceGAN: Discriminative Classification Framework for Rare Class Identification for Longitudinal Data with Weak Prior

Surajit Chakrabarty, Rukma Talwadker, Tridib Mukherjee

TL;DR

ScarceGAN addresses the challenge of identifying extremely rare events in multi-dimensional longitudinal telemetry with tiny and weak priors. It extends semi-supervised GANs by partitioning negative samples into multiple sub-classes, introducing a leeway Unknown class, and employing a complementary generator to learn in the discriminator's complement space, all while leveraging unlabelled data. The framework delivers high recall for scarce positives (over 85%) and outperforms state-of-the-art methods on both synthetic skill-game data and the KDDCUP99 intrusion dataset, including scenarios with imbalances as small as 0.09%. Through longitudinal feature synthesis and a tailored loss design for supervised and unsupervised paths, ScarceGAN establishes a robust benchmark for rare-class identification in longitudinal telemetry and related domains.

Abstract

This paper introduces ScarceGAN which focuses on identification of extremely rare or scarce samples from multi-dimensional longitudinal telemetry data with small and weak label prior. We specifically address: (i) severe scarcity in positive class, stemming from both underlying organic skew in the data, as well as extremely limited labels; (ii) multi-class nature of the negative samples, with uneven density distributions and partially overlapping feature distributions; and (iii) massively unlabelled data leading to tiny and weak prior on both positive and negative classes, and possibility of unseen or unknown behavior in the unlabelled set, especially in the negative class. Although related to PU learning problems, we contend that knowledge (or lack of it) on the negative class can be leveraged to learn the compliment of it (i.e., the positive class) better in a semi-supervised manner. To this effect, ScarceGAN re-formulates semi-supervised GAN by accommodating weakly labelled multi-class negative samples and the available positive samples. It relaxes the supervised discriminator's constraint on exact differentiation between negative samples by introducing a 'leeway' term for samples with noisy prior. We propose modifications to the cost objectives of discriminator, in supervised and unsupervised path as well as that of the generator. For identifying risky players in skill gaming, this formulation in whole gives us a recall of over 85% (~60% jump over vanilla semi-supervised GAN) on our scarce class with very minimal verbosity in the unknown space. Further ScarceGAN outperforms the recall benchmarks established by recent GAN based specialized models for the positive imbalanced class identification and establishes a new benchmark in identifying one of rare attack classes (0.09%) in the intrusion dataset from the KDDCUP99 challenge.

ScarceGAN: Discriminative Classification Framework for Rare Class Identification for Longitudinal Data with Weak Prior

TL;DR

ScarceGAN addresses the challenge of identifying extremely rare events in multi-dimensional longitudinal telemetry with tiny and weak priors. It extends semi-supervised GANs by partitioning negative samples into multiple sub-classes, introducing a leeway Unknown class, and employing a complementary generator to learn in the discriminator's complement space, all while leveraging unlabelled data. The framework delivers high recall for scarce positives (over 85%) and outperforms state-of-the-art methods on both synthetic skill-game data and the KDDCUP99 intrusion dataset, including scenarios with imbalances as small as 0.09%. Through longitudinal feature synthesis and a tailored loss design for supervised and unsupervised paths, ScarceGAN establishes a robust benchmark for rare-class identification in longitudinal telemetry and related domains.

Abstract

This paper introduces ScarceGAN which focuses on identification of extremely rare or scarce samples from multi-dimensional longitudinal telemetry data with small and weak label prior. We specifically address: (i) severe scarcity in positive class, stemming from both underlying organic skew in the data, as well as extremely limited labels; (ii) multi-class nature of the negative samples, with uneven density distributions and partially overlapping feature distributions; and (iii) massively unlabelled data leading to tiny and weak prior on both positive and negative classes, and possibility of unseen or unknown behavior in the unlabelled set, especially in the negative class. Although related to PU learning problems, we contend that knowledge (or lack of it) on the negative class can be leveraged to learn the compliment of it (i.e., the positive class) better in a semi-supervised manner. To this effect, ScarceGAN re-formulates semi-supervised GAN by accommodating weakly labelled multi-class negative samples and the available positive samples. It relaxes the supervised discriminator's constraint on exact differentiation between negative samples by introducing a 'leeway' term for samples with noisy prior. We propose modifications to the cost objectives of discriminator, in supervised and unsupervised path as well as that of the generator. For identifying risky players in skill gaming, this formulation in whole gives us a recall of over 85% (~60% jump over vanilla semi-supervised GAN) on our scarce class with very minimal verbosity in the unknown space. Further ScarceGAN outperforms the recall benchmarks established by recent GAN based specialized models for the positive imbalanced class identification and establishes a new benchmark in identifying one of rare attack classes (0.09%) in the intrusion dataset from the KDDCUP99 challenge.
Paper Structure (22 sections, 10 equations, 7 figures, 6 tables)

This paper contains 22 sections, 10 equations, 7 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. ScarceGAN Formulation
  4. Background: GAN and SSGAN
  5. SSGAN Formulation
  6. ScarceGAN Formulation
  7. ScarceGAN Discriminator
  8. Supervised Path:
  9. Unsupervised Path:
  10. ScarceGAN Generator
  11. Handling longitudinal data
  12. Feature Synthesis:
  13. Empirical Evaluation
  14. Datasets:
  15. Model Training and Test Set Description:
  16. ...and 7 more sections

Figures (7)

  • Figure 1: (A) no separability between positive-negative samples. (B) poorly augmented positive samples with limited positive data
  • Figure 2: Mix of Overlapping and Non-Overlapping patterns in negative class samples
  • Figure 3: ScarceGAN Architecture
  • Figure 4: Feature Time Series Modelling using Customized Prophet: dots - actual samples, blue line - predicted values, red line - trend line
  • Figure 5: AAE based positive detection: MSE and Likelihood thresholds cannot escape verbosity. Highly overlapping distribution between positives and negatives in our case.
  • ...and 2 more figures