Publications

CAST: Causal Anchored Simplex Transport for Distribution-Valued Time Series

NeurIPS 2026

Jiecheng Lu, Jieqi Di, Runhua Wu, Yuwei Zhou

Introduces CAST, a successor-local operator for forecasting distribution-valued time series that retrieves empirical successors from causal context, anchors them to persistence, and applies bounded local transport on the probability simplex.

PDF arXiv

HyperMLP: An Integrated Perspective for Sequence Modeling

ICML 2026

Jiecheng Lu, Shihao Yang

Presents an integrated dynamic-MLP perspective on sequence modeling, reinterpreting attention heads through context-instantiated MLP computation and learnable sequence-space mixing.

PDF Conference page

StretchTime: Adaptive Time Series Forecasting via Symplectic Attention

ICML 2026

Yubin Kim, Viresh Pati, Jevon Twitty, Vinh Pham, Shihao Yang, Jiecheng Lu*

Introduces adaptive time series forecasting via symplectic attention, developed through mentored undergraduate research with Jiecheng Lu as corresponding author.

PDF Conference page

Free Energy Mixer

ICLR 2026

Jiecheng Lu, Shihao Yang

Introduces Free Energy Mixer (FEM), which interprets (q,k) attention scores as a prior and performs a log-sum-exp free-energy readout to reweight values at the channel level, enabling a smooth transition from mean aggregation to selective channel-wise retrieval without increasing asymptotic complexity.

OpenReview Code Poster

ZeroS: Zero‑Sum Linear Attention for Efficient Transformers

NeurIPS 2025

Jiecheng Lu, Xu Han, Yan Sun, Viresh Pati, Yubin Kim, Siddhartha Somani, Shihao Yang

Spotlight

Introduces Zero‑Sum Linear Attention (ZeroS), which removes the uniform zero‑order term and reweights residuals to enable stable positive/negative attention weights, allowing contrastive operations within a single layer while retaining O(N) complexity.

PDF Proceedings Code Poster

Linear Transformers as VAR Models: Aligning Autoregressive Attention Mechanisms with Autoregressive Forecasting

ICML 2025

Jiecheng Lu, Shihao Yang

Shows that a linear attention layer can be interpreted as a dynamic VAR; proposes SAMoVAR to realign multi‑layer Transformers with autoregressive forecasting for improved interpretability and accuracy.

PDF Proceedings arXiv Code Poster

WAVE: Weighted Autoregressive Varying Gate for Time Series Forecasting

ICML 2025

Jiecheng Lu, Xu Han, Yan Sun, Shihao Yang

Adds ARMA structure to autoregressive attention via a weighted varying gate, decoupling long‑range and local effects and improving TSF quality without increasing asymptotic complexity.

PDF Proceedings arXiv Code Poster

In‑context Time Series Predictor

ICLR 2025

Jiecheng Lu, Yan Sun, Shihao Yang

Reformulates TSF as in‑context learning by constructing tokens of (lookback, future) task pairs, enabling Transformers to adapt predictors from context without parameter updates.

PDF OpenReview arXiv Code Poster Slides

CATS: Enhancing Multivariate Time Series Forecasting by Constructing Auxiliary Time Series as Exogenous Variables

ICML 2024

Jiecheng Lu, Xu Han, Yan Sun, Shihao Yang

Constructs Auxiliary Time Series (ATS) as exogenous inputs to capture inter‑series relations; identifies continuity, sparsity, and variability principles; improves multivariate TSF even with simple predictors.

PDF Proceedings arXiv Code Poster

ARM: Refining Multivariate Forecasting with Adaptive Temporal‑Contextual Learning

ICLR 2024

Jiecheng Lu, Xu Han, Shihao Yang

Presents ARM with AUEL, Random Dropping, and multi‑kernel local smoothing to better capture series‑wise patterns and inter‑series dependencies for long‑term multivariate TSF.

PDF Proceedings arXiv Code Poster