MLx Representation Learning & Generative AI 25

This module delves into conformal prediction and its synergy with Bayesian methods, focusing on uncertainty quantification in machine learning. Students will learn the fundamentals of conformal prediction, its applications in diverse fields, and how it enhances Bayesian predictions. The course emphasizes the significance of data splitting and addresses practical challenges in model calibration, equipping learners with essential skills for effective decision-making in uncertain environments.

1. 1. Talk
2. 2. Q&A

This module delves into the critical aspects of AI safety, emphasizing the challenges of bias, deception, and manipulation in machine learning systems. Students will learn interpretability techniques and engage in hands-on exercises to analyze AI behavior, ultimately equipping them with the knowledge to enhance safety measures in AI applications.

1. 1. Talk

This module delves into the creation of generalist AI agents and the ARC robotics framework, emphasizing the integration of human psychology theories and advanced learning techniques. Students will explore the principles of adaptive learning, imitation learning, and the significance of physics grounding in robotics, equipping them with the knowledge to innovate in AI and robotics applications.

1. 1. Talk
2. 2. Q&A

In this module, students will delve into the architecture of multimodal models, focusing on the integration of vision, audio, and language encoders. They will explore cutting-edge advancements like quantization and reasoning models, while gaining practical insights through demonstrations of multimodal applications. This knowledge is essential for developing efficient and effective AI systems that leverage multiple data types.

1. 1. Talk

This module delves into the evolution of AI methodologies, emphasizing the transition to deep learning and foundation models. Students will learn to apply language models for uncertainty quantification in data science, enhancing decision-making in critical fields. The course also covers advanced practices in large language models, including parameter selection and multimodal data integration, equipping learners with the knowledge to optimize AI performance in real-world applications.

1. 1. Talk
2. 2. Talk
3. 3. Q&A

In this module, students will delve into the intricacies of evaluating machine learning models for tabular and time series data. They will learn about the effectiveness of AutoML systems, the importance of evaluation metrics, and hyperparameter optimization. Additionally, students will explore the capabilities of agents in medical and forecasting contexts, gaining insights into tools like AutoGalon and foundational models for enhanced data analysis.

1. 1. Talk
2. 2. Q&A

This module delves into the intersection of generative AI and representation learning in robotics, equipping students with knowledge on data-efficient learning methods. Students will explore the integration of pretrained language models, graph neural networks, and innovative techniques to overcome data scarcity challenges. The focus on aligning image and action spaces will prepare learners to enhance robot adaptability and task performance in real-world environments.

1. 1. Talk
2. 2. Q&A

This module delves into the evolution and application of denoising techniques and diffusion models in machine learning and computational imaging. Students will explore the significance of denoisers in image processing, the challenges of non-Gaussian distributions, and the intricacies of 3D data modeling. By understanding these concepts, learners will enhance their ability to develop robust generative models and tackle real-world imaging problems.

1. 1. Talk
2. 2. Q&A

In this module, students will delve into the intricacies of foundation models for time series data, focusing on advanced forecasting techniques and the role of tokenization. They will learn to navigate challenges in multivariate forecasting and anomaly detection, while gaining insights into innovative models like Kronos. This knowledge is essential for making accurate probabilistic predictions in various applications.

1. 1. Talk
2. 2. Q&A

In this module, students will delve into the evolution and mechanics of generative AI, focusing on image and video generation techniques. They will explore foundational concepts such as representation learning, variational autoencoders, and diffusion models, while addressing the challenges of creating realistic outputs. This knowledge is crucial for developing innovative AI applications in various fields.

1. 1. Talk
2. 2. Q&A

This module delves into the critical role of representation learning in reinforcement learning, emphasizing its impact on sample efficiency and robustness. Students will explore safety constraints in motion primitives, optimization techniques for diffusion models, and the challenges of transferring visual data from simulation to real-world applications. By understanding these concepts, learners will be equipped to improve AI performance in robotics and industrial settings.

1. 1. Talk
2. 2. Q&A

This module delves into the intricacies of deploying AI models on embedded and mobile devices, emphasizing the use of PyTorch and the Executorch framework. Students will learn about model compilation for hardware accelerators, the challenges of on-device training, and the impact of quantization on performance. By understanding these concepts, learners will be equipped to tackle real-world deployment challenges while ensuring efficiency and privacy.

1. 1. Talk
2. 2. Q&A

This module delves into the intricacies of multi-agent cooperation in AI, emphasizing the synergy between human knowledge and AI systems. Students will explore knowledge extraction from texts, the representation of that knowledge in code, and the challenges of effective human-AI collaboration. By understanding these dynamics, learners will be equipped to enhance AI interactions and develop future personal assistant technologies.

1. 1. Talk
2. 2. Q&A

This module delves into the evolution of foundation models in computer vision, emphasizing their limitations and the need for multimodal understanding. Students will explore frame-based attention mechanisms, knowledge distillation, and the integration of 2D and 3D tracking models, equipping them with insights into the complexities of model training and data selection.

1. 1. Talk
2. 2. Q&A

In this module, students will delve into the intricacies of training open language models, focusing on transparency, reproducibility, and effective data curation. They will explore advanced alignment techniques, including reinforcement learning from human feedback, and understand the challenges of maintaining model performance. This knowledge is crucial for developing robust AI systems that adhere to open scientific practices.

1. 1. Talk
2. 2. Q&A