Machine Learning Researcher, Vulcan

Job Description

We are seeking a skilled and experienced Machine Learning (ML) Researcher to contribute to the development of cutting-edge safety and security solutions for ML systems, with a strong focus on large language & multi-modal models (LLMs) and their applications. The ideal candidate will have hands-on experience building and deploying LLMs in production environments, combined with a passion for addressing challenges related to adversarial attacks, model robustness, data privacy, and compliance.

我們正在尋找一位具備豐富經驗的機器學習研究員，專注於研究最前沿的ML系統安全與防護解決方案，特別是大型語言模型與多模態模型及其應用領域。該職位應具備ML的研究與開發經驗，並對於對抗式攻擊、模型穩健性、數據隱私與合規等挑戰充滿熱情，致力於推動更安全、更可靠的AI解決方案。

Vulcan: https://vulcanlab.ai/

Cymetrics: https://cymetrics.io/zh-tw/products/ai-redteam

OneDegree Tech Blog: https://medium.com/onedegree-tech-blog

-

How to apply

It will help us process your applications faster
*Please apply by English CV, thank you.

-

Responsibilities

Research and Development:

• Conduct original research on ML safety and security topics, including adversarial robustness, LLM interpretability, bias detection, and secure training protocols.針對 ML 安全與防護 進行原創性研究，包括 對抗式攻擊防禦、LLM 可解釋性、偏見偵測 以及 安全訓練協議。
• Develop state-of-the-art techniques to identify and mitigate risks specific to LLMs, such as prompt injection, data leakage, and unintended outputs.開發最先進技術，識別並緩解 LLM 風險，如 Prompt 注入攻擊、數據洩露、非預期輸出 等問題。
• Explore scalable approaches for ensuring model safety, fairness, and reliability in production environments.
  探索可擴展的方法，以確保 模型的安全性、公平性與穩定性，並能適用於生產環境。

Practical Development and Deployment:

• Design, develop, and deploy large language models (LLMs) for production use cases, ensuring they meet high standards of performance, reliability, and safety.
  設計、開發並部署 大型語言模型，確保其在生產環境中具備高效能、可靠性與安全性。
• Optimize LLMs for resource efficiency and integrate safety and security features into deployment pipelines.
  優化 LLM 的資源使用效率，並將安全防護功能整合至部署流程。
• Implement monitoring tools to detect and address real-world threats to deployed ML systems, including LLMs.
  實作監控工具，偵測與應對 LLM 及 ML 系統的潛在安全威脅。

Threat Analysis and Risk Mitigation:

• Identify vulnerabilities and attack vectors in ML systems, particularly in LLM-based applications.
  識別 ML 系統漏洞與攻擊向量，特別是基於 LLM 的應用。
• Develop tools and strategies for protecting LLM systems from adversarial attacks, data poisoning, and unintended behaviors.
  開發防禦工具與策略，防範 對抗式攻擊、數據投毒 及 非預期行為。
• Build frameworks to evaluate the safety and security of LLMs under various operational scenarios.
  建立安全性評估框架，測試 LLM 在不同運行場景下的安全性與穩定性。

Collaboration and Integration:

• Collaborate with cross-functional teams, including engineers, product managers, and domain experts, to align research efforts with business goals.
  與 工程師、產品經理、領域專家 合作，確保研究成果符合業務目標。
• Work closely with DevOps teams to integrate research outcomes into scalable and reliable LLM deployment workflows.
  與 DevOps 團隊 緊密合作，將研究成果整合至 LLM 部署流程，確保其可擴展性與可靠性。

Compliance and Ethics:

• Ensure LLM deployments comply with relevant safety, security, and data privacy regulations.
  確保 LLM 部署符合資安、隱私與法規要求。
• Advocate for ethical and transparent AI practices in product development.
  推動 AI 倫理與透明度，確保 AI 產品開發符合公平性與合規性標準。

Thought Leadership:

• Publish research findings in leading journals and conferences to contribute to the advancement of ML safety and security.
  發表研究成果，參與頂尖學術期刊與 AI 安全會議，推動 ML 安全領域的發展。
• Represent the organization in academic and industry forums focused on AI safety and security.
  代表公司參與 AI 安全與資安相關論壇，提升業界影響力。

Requirements

Education Background:
Bachelor's, Master's, or Ph.D. in Computer Science, Artificial Intelligence, Machine Learning, Cybersecurity, or a related field. Equivalent industry experience will also be considered.
計算機科學、人工智慧、機器學習、資安或相關領域的學士、碩士或博士學位。具備同等業界經驗者亦可。

Technical Skills:

• Strong programming expertise in Python and experience with ML frameworks such as PyTorch, TensorFlow, or similar.
  精通 Python，並具備 PyTorch、TensorFlow 或類似機器學習框架的開發經驗。
• Proven experience building, fine-tuning, and deploying LLMs and other NLP models (e.g., BERT) in production environments.
  具備大型語言模型與自然語言處理模型（如 BERT）的開發、微調與生產環境部署經驗。
• In-depth knowledge of adversarial ML, differential privacy, and secure training practices.
  熟悉對抗式機器學習、差分隱私與安全訓練技術。
• Experience with MLOps tools (e.g., Kubeflow, MLflow .. etc) for deploying and managing ML models in production.
  具備 MLOps 工具（如 Kubeflow、MLflow）操作經驗，能有效管理與部署 ML 模型至生產環境。

Experience:

• Hands-on experience developing, deploying, and optimizing large-scale ML models, particularly LLMs, for real-world applications.
  具備大型 ML 模型（特別是 LLMs）在實際應用場景中的開發、部署與優化經驗。
• A proven track record of addressing security and safety concerns in deployed ML systems.
  成功處理 ML 系統安全與穩定性問題的經驗，包括風險分析與安全性提升。
• Experience with data preprocessing, model evaluation, and performance tuning for LLMs in production.
  熟悉 LLMs 的數據預處理、模型評估與效能調校，確保模型在生產環境中的最佳運行。
• Experience in identifying emerging ML and AI threats.
  具備識別 AI/ML 領域新興威脅的經驗，可主動發掘與應對潛在風險。

Soft Skills:

• Strong problem-solving and critical-thinking abilities.
  優秀的問題解決與批判性思維能力。
• Excellent communication skills, with the ability to convey technical concepts to diverse audiences.
  出色的溝通能力，能夠向不同背景的受眾清楚表達技術概念。
• Ability to write and speak in English fluently.
  流利的英文書寫與口語表達能力。
• Passion for developing robust, secure, and ethical AI systems.
  熱衷於開發安全、穩健且符合倫理的 AI 系統。

Nice-to-Have:

• Experience with AI model interpretability and explainability techniques.
  熟悉 AI 模型可解釋性與可解釋 AI 技術。
• Knowledge of federated learning, differential privacy, and secure AI training methodologies.
  了解聯邦學習、差分隱私、安全 AI 訓練方法。
• Background in AI compliance and auditing.
  具備 AI 合規性與審計相關經驗。

加分條件（Nice-to-Have）

• Familiarity with prompt engineering and LLM evaluation methodologies.
  熟悉 Prompt Engineering 與 LLM 評估方法。
• Knowledge of regulatory frameworks (e.g., GDPR, CCPA, AI Act) and secure software development practices.
  了解 AI 相關法規（如 GDPR、CCPA、AI Act），並具備安全軟體開發實踐經驗。
• Experience working with interdisciplinary teams (e.g., legal, compliance, or policy).
  曾與跨領域團隊（如法務、合規、政策制定）合作，解決 AI 安全與合規問題的經驗。