Advanced Positional Embeddings

Background: Efficiently encoding temporal information across irregularly sampled data points is crucial for maintaining accuracy in time series models like the P-STMAE.

Question / Future Work: Future work should investigate advanced positional embedding techniques for irregular time series, such as Attention with Linear Biases (ALiBi) or Rotary Position Embedding (RoPE), to better capture relative temporal relationships without relying solely on explicit positional encodings.