Remote | Machine Learning Research Scientist — $95–$115/hour

24-Mag • New York, NY

20h • Remote

About The Position

We are sharing a specialised consulting opportunity for experienced machine learning researchers with hands-on expertise training and improving deep learning models end-to-end across computer vision and language. This role supports advanced empirical machine learning research across model training, efficiency, robustness, multimodal systems, and post-training. Selected researchers will work on well-scoped but open-ended technical problems involving image models, language models, adversarial robustness, model compression, multilingual learning, and efficient training under constrained data and compute budgets.

Requirements

• At least 3 years of machine learning research experience, including qualifying PhD research
• Hands-on experience training deep learning models end-to-end
• Strong proficiency with PyTorch, JAX, TensorFlow, or comparable machine learning frameworks
• Deep expertise in at least one relevant research area such as adversarial robustness, computer vision, generative modelling, LLM post-training, or multilingual pre-training
• Experience designing and running rigorous empirical experiments
• Strong understanding of optimisation, model evaluation, and experimental methodology
• Ability to diagnose complex model-training and performance issues
• Strong technical writing and research communication skills
• A degree in computer science, machine learning, artificial intelligence, mathematics, statistics, engineering, or a related technical field is highly relevant
• PhD research in machine learning or a closely related area may count toward the professional experience requirement
• Candidates may also demonstrate equivalent research strength through significant industry work, publications, or impactful open-source contributions
• A strong academic, industry, or independent research track record is particularly valuable

Nice To Haves

• Experience with scaling laws or training-efficiency research
• Background in curriculum learning or data ordering
• Experience building machine learning benchmarks
• Knowledge of benchmark contamination detection and prevention
• Familiarity with statistically rigorous model comparison
• Experience with uncertainty estimation or model calibration
• Expertise in synthetic data or data augmentation
• Publications in recognised machine learning or AI venues
• Experience at a major AI, technology, or research organisation
• Significant open-source machine learning contributions

Responsibilities

• Train image classifiers and generative image models from scratch
• Fine-tune and post-train open-weight language models
• Design and execute empirical machine learning experiments
• Diagnose optimisation, convergence, data-quality, and training-stability issues
• Develop approaches that maximise performance under limited data, compute, or model-size budgets
• Train image classifiers for challenging recognition tasks
• Develop models for fine-grained recognition with limited examples
• Train diffusion models, GANs, VAEs, flow-based models, or comparable generative architectures
• Evaluate generative models using metrics such as FID
• Improve sample quality while controlling training cost and parameter count
• Develop models that remain reliable under adversarial inputs
• Apply adversarial training approaches such as PGD-based training or TRADES
• Evaluate robust accuracy under established threat models
• Investigate robustness–accuracy trade-offs and robust overfitting
• Apply quantisation, pruning, knowledge distillation, and related model-compression techniques
• Optimise models for strict memory, size, or latency constraints
• Conduct supervised fine-tuning and preference optimisation of open-weight language models
• Work with methods such as DPO, RLHF, or RLAIF where relevant
• Develop training datasets using synthetic generation, weak supervision, noisy supervision, or rejection sampling
• Improve multi-turn conversational behaviour including resistance to persuasion and sycophancy
• Develop approaches for calibrated confidence and appropriate response to corrections
• Modify targeted behaviours while preserving broader model capabilities
• Train multilingual or low-resource language models
• Develop tokenisation strategies across diverse scripts and language families
• Address highly imbalanced multilingual training datasets
• Explore sampling strategies and cross-lingual transfer
• Improve model performance in data-constrained language settings

Benefits

• Flexible project-based work
• Competitive hourly compensation
• Competitive rates between $95–$115 per hour depending on expertise and project scope
• Weekly payments via Stripe or Wise
• Projects may be extended, shortened, or adjusted depending on scope and performance

What This Job Offers

• Career Level: Senior