Beyond Imputation: Missingness in Wearable Tech with Maxwell Xu скачать в хорошем качестве

Beyond Imputation: Missingness in Wearable Tech with Maxwell Xu

The desire for deeper insights into intimate health is growing, driven in part by the rapid development of wearable technology. Or is it the other way around? Either way, the wearable market is expanding quickly, allowing us to gather increasingly rich data about our daily health and well-being. But this raises important questions: What challenges come with collecting this data? And perhaps even more interesting—what happens when that data is missing? In this episode, we speak with a lead Google researcher, Maxwell Xu, about his paper, LSM-2: Learning from Incomplete Wearable Sensor Data, a next-generation foundation model for wearable sensor data that can learn directly from incomplete and fragmented signals. Unlike traditional self-supervised models that require clean, fully observed data, LSM-2 uses a new method called Adaptive and Inherited Masking (AIM) to handle real-world missingness without data imputation. Full Paper: https://arxiv.org/abs/2506.05321 Chapters 0:00 Journey to ML 2:40 The Problem Space 6:09 Prior Work in Missingness 9:40 Novel Approach 12:00 Introducing Bias with Imputation 12:53 Building on LSM-1 15:21 Why Sample Regularity Matters 18:07 How It Works 20:09 Experiments 21:26 Surprising Results 23:00 Implications 25:57 Limitations 27:07 Wearable Tech & Healthcare #wearabletech #ai #techpodcast #missingdata #futureoftech #aiinhealthcare #futureofhealthcare