Description and Requirements
About the Team
We focus on emerging paradigms that may reshape the next generation of AI. Beyond mainstream autoregressive large models, we are placing early bets on diffusion language models, hardware-software co-design, machine-oriented structured decision models, latent-space reasoning, and world models / Physical AI. Our goal is to build core capabilities for long-term technical competitiveness, with on-device products and on-device solutions as the ultimate landing scenarios.
Key Responsibilities
1. Frontier Paradigm Research: Within the team's overall technical roadmap, focus on one or more of the following areas based on your background, and independently own research on specific sub-topics, including but not limited to:
- Diffusion Language Models: Research modeling, training, and parallel decoding methods for non-autoregressive and diffusion-based language models to improve generation quality and inference speed.
- Hardware-Software Co-Design: Research model architectures and training/inference solutions tailored to on-device hardware, including low-bit training and inference, sparsity, and operator and memory-access optimization.
- Structured Decision Models: Research lightweight decision models built for machines (e.g., Jev-like models), delivering typed outputs, confidence calibration, and ultra-low-latency inference.
- Latent-Space Reasoning: Research methods that perform reasoning in continuous latent states, reducing the token overhead of explicit chain-of-thought and improving reasoning efficiency.
- World Models / Physical AI: Research world models that represent, predict, and plan in the physical world, and explore their applications in on-device perception.
2. Algorithm Development & Experimentation: Own method design, model training, experimental validation, and performance analysis for your research topics. Reproduce and improve on state-of-the-art methods, and draw reliable experimental conclusions.
3. Data & Evaluation: Help build datasets and evaluation benchmarks in your area, providing quantifiable ways to validate emerging paradigms.
4. Research-to-Product Transfer: Work with engineering and on-device system teams to turn research prototypes into on-device products and solutions.
5. Research Output: Keep up with frontier advances in the field and produce high-quality publications, open-source projects, and technical reports.
Qualifications
1. Master's degree or above in Computer Science, Artificial Intelligence, Electrical Engineering, Mathematics, or a related field Ph.D. preferred. 3+ years of relevant research experience with a Master's degree, or a Ph.D. requirements may be relaxed for exceptional candidates.
2. Solid grasp of the fundamentals of deep learning and large models familiarity with mainstream architectures such as Transformer strong skills in model training and experimental design.
3. In-depth research or development experience in at least one of the following areas, with the ability to independently design, validate, and optimize key algorithms:
- Diffusion language models, non-autoregressive generation
- Model architecture design, hardware-software co-design, efficient training and inference
- Structured outputs, confidence calibration, lightweight decision models
- Latent-space reasoning, LLM reasoning
- World models, self-supervised representation learning, embodied AI
4. Strong research fundamentals and analytical skills able to independently analyze and solve key technical problems, and comfortable with the uncertainty of frontier exploration.
5. Proficiency in Python, PyTorch, and related tools, with strong coding skills and hands-on model training experience.
Preferred Qualifications
1. High-quality publications at venues such as NeurIPS, ICML, ICLR, CVPR, ICCV/ECCV, ACL, and EMNLP, or contributions to influential open-source projects.
2. Experience in pre-training or training from scratch novel models such as diffusion language models or world models.
3. Experience in large-scale distributed training, model quantization and compression, or on-device deployment.
4. Experience with on-device hardware such as NPU/DSP, or hardware-software co-design projects.
