Conference Papers
This page contains peer-reviewed conference publications focusing on machine learning, computer vision, and cross-modal systems.
2024
“Tiny Contrastive Learning for Resource-Constrained Environments”
Authors: Stephen Mander, et al.
Conference: International Conference on Learning Representations (ICLR) 2024
Status: Published
Links: PDF | Code | Project Page
Abstract: This paper presents novel approaches for implementing contrastive learning in resource-constrained environments. We develop efficient algorithms that maintain representation quality while significantly reducing computational requirements, enabling deployment on edge devices and mobile platforms.
Key Contributions:
Novel lightweight contrastive learning architectures
Resource-aware training strategies for edge deployment
Comprehensive evaluation across multiple resource-constrained scenarios
Open-source framework for efficient contrastive learning
Research Impact:
Applications: Mobile AI, edge computing, IoT systems
Technical Innovation: 60% reduction in computational requirements with minimal performance loss
Community Impact: Framework adopted by multiple research groups for edge AI applications
“BertScore Evaluation Framework for Semantic Similarity Assessment”
Authors: Stephen Mander, et al.
Conference: Natural Language Processing and Information Systems (NLDB) 2024
Status: Published
Links: PDF | Evaluation Code
Abstract: We present a comprehensive evaluation framework for BertScore-based semantic similarity metrics, analyzing their effectiveness across diverse text domains and proposing improvements for cross-lingual applications.
Key Contributions:
Systematic evaluation of BertScore variants across multiple domains
Novel cross-lingual extensions for semantic similarity assessment
Comprehensive benchmark dataset for semantic similarity evaluation
Practical guidelines for BertScore deployment in production systems
Research Impact:
Benchmark Creation: New evaluation standard for semantic similarity metrics
Cross-lingual Innovation: First comprehensive study of BertScore in multilingual contexts
Practical Impact: Guidelines adopted by industry practitioners for NLP system evaluation
2023
“PICTAR: Probe Informed Contrastive Training for Adversarial Robustness”
Authors: Stephen Mander, et al.
Conference: Computer Vision and Pattern Recognition (CVPR) Workshop 2023
Status: Published
Abstract: PICTAR introduces a novel training methodology that combines probing techniques with contrastive learning to improve adversarial robustness in computer vision models. Our approach achieves state-of-the-art robustness while maintaining competitive clean accuracy.
Key Contributions:
Novel probe-informed contrastive training methodology
Theoretical analysis of robustness-accuracy trade-offs
Comprehensive evaluation across multiple attack scenarios
Open-source adversarial training and visualization framework
Research Impact:
Security Applications: Enhanced robustness for safety-critical vision systems
Theoretical Insights: New understanding of probe-guided training dynamics
Practical Tools: Visualization framework for adversarial training analysis
“Improving PGN Attacks with Unsupervised Training Strategies”
Authors: Stephen Mander, et al.
Conference: International Conference on Machine Learning (ICML) Workshop 2023
Status: Published
Links: PDF | Attack Framework
Abstract: This work explores unsupervised training strategies for enhancing Projected Gradient Descent (PGD) attacks, demonstrating significant improvements in attack success rates across multiple model architectures and defense mechanisms.
Key Contributions:
Novel unsupervised training approaches for adversarial attacks
Systematic analysis of attack transferability across model families
Comprehensive evaluation framework for attack effectiveness
Open-source implementation of enhanced attack algorithms
Research Impact:
Security Research: Advanced understanding of adversarial vulnerabilities
Defense Development: Insights enabling more robust defense mechanisms
Evaluation Standards: New benchmarks for adversarial attack assessment
2022
“8-Bit LSA: Efficient Latent Semantic Analysis Implementation”
Authors: Stephen Mander, et al.
Conference: International Conference on Computational Linguistics (COLING) 2022
Status: Published
Links: PDF | Implementation
Abstract: We present an efficient 8-bit implementation of Latent Semantic Analysis that reduces memory requirements by 75% while maintaining semantic representation quality, enabling deployment on resource-constrained systems.
Key Contributions:
Novel 8-bit quantization strategy for LSA models
Memory-efficient implementation with minimal quality loss
Comprehensive evaluation across multiple text corpora
Open-source framework for efficient semantic analysis
Research Impact:
Resource Efficiency: 75% reduction in memory requirements
Deployment Innovation: Enables LSA on mobile and embedded systems
Community Adoption: Implementation used by multiple NLP research groups
Research Themes & Ongoing Work
Cross-Modal Learning Systems
My conference publications demonstrate a consistent focus on advancing cross-modal learning through:
Contrastive Learning Innovation: Developing efficient methods for multi-modal representation learning
Adversarial Robustness: Enhancing security and reliability of vision-language models
Resource Optimization: Creating practical solutions for real-world deployment constraints
Future Conference Submissions
Upcoming Conferences (2025):
NeurIPS 2025: Scaling laws in contrastive autoencoders
ICLR 2025: Multi-dimensional CLIP training methodologies
CVPR 2025: Advanced adversarial robustness techniques
For related journal publications, see Journal Articles. For work in progress, see Planned Work.
Last updated: Sep 16, 2025