https://mwxinnn.github.io/Recent content on Wenxin (Wendy) Ma | Computer Vision & Medical AIHugoen-usFri, 11 Oct 2024 00:00:00 +0000https://mwxinnn.github.io/tips/ml/loss/kl/Fri, 11 Oct 2024 00:00:00 +0000https://mwxinnn.github.io/tips/ml/loss/kl/https://mwxinnn.github.io/tips/ml/optimization/catastrophes/Thu, 12 Sep 2024 00:00:00 +0000https://mwxinnn.github.io/tips/ml/optimization/catastrophes/Vanishing/Exploding Gradients Catastrophic interference Catastrophic interference, also known as catastrophic forgetting, is the tendency of an artificial neural network to abruptly and drastically forget previously learned information upon learning new information. When a network learns a new task, it modifies its weights to reduce the error for that particular task. This modification can dramatically alter the knowledge representation of prior tasks, leading to the “forgetting” phenomenon. One example of the catastrophic forgetting problem is when training a model on the MNIST classification task with digits 0-4 and testing it on all digits 0-9.https://mwxinnn.github.io/tips/ml/optimization/optimizer/prerequisite/Mon, 12 Aug 2024 00:00:00 +0000https://mwxinnn.github.io/tips/ml/optimization/optimizer/prerequisite/Momentum Momentum is a technique used in optimization to accelerate the convergence of gradient descent by smoothing out the oscillations in the optimization path. The core idea of momentum is to accumulate a moving average of past gradients and use this average to update the parameters, rather than relying solely on the current gradient. It helps in pushing the parameters in the consistent direction of descent by giving more weight to gradients from previous steps.https://mwxinnn.github.io/tips/ml/optimization/overfitting/Mon, 12 Aug 2024 00:00:00 +0000https://mwxinnn.github.io/tips/ml/optimization/overfitting/https://mwxinnn.github.io/tips/ml/optimization/scheduler/Fri, 09 Aug 2024 00:00:00 +0000https://mwxinnn.github.io/tips/ml/optimization/scheduler/https://mwxinnn.github.io/tips/git/branchmanagement/Thu, 08 Aug 2024 00:00:00 +0000https://mwxinnn.github.io/tips/git/branchmanagement/Basic Understandings A branch in Git is essentially a pointer to a specific commit in the repository’s history. The primary branch, usually called main, represents the production-ready code. Additional branches are used to develop features, fix bugs, or experiment without affecting the main branch. main: Default Local Branch created with git init. Prior to Git version 2.28 (released in 2020), the default branch was named master. origin/main: When you clone a repository using git clone <repository_url>, Git automatically sets up a remote named origin.https://mwxinnn.github.io/tips/git/conflicthandling/Thu, 08 Aug 2024 00:00:00 +0000https://mwxinnn.github.io/tips/git/conflicthandling/When attempting to merge a branch that results in conflicts, Git will output which files are in conflict. Checking the status of the repository, conflicted files will be marked with “both modified”. One need to manually edit the file to resolve conflicts by choosing which changes to keep or combining them. To see conflict markers in files: git diff After resolving the conflicts in the file, add the resolved file to the staging area and complete the merge by committing the changes.https://mwxinnn.github.io/tips/git/rebase/Thu, 08 Aug 2024 00:00:00 +0000https://mwxinnn.github.io/tips/git/rebase/Rebase Merge combines changes from one branch into another while preserving the history of both branches. It creates a new commit (a “merge commit”) that combines the changes from the two branches. The merge commit has two parent commits (one from each branch being merged). It maintains a complete history of changes and the context in which they were made. Rebase reapplies commits from one branch on top of another, creating a linear history.https://mwxinnn.github.io/details/education/Mon, 01 Jan 0001 00:00:00 +0000https://mwxinnn.github.io/details/education/PhD student, Computer Science — The University of Texas at Austin First year Advised by Prof. Philipp Krähenbühl. Visiting Scholar — Whiting School of Engineering, Johns Hopkins University Joined Apr 2025 Research with Prof. Alan Yuille and Ph.D. candidate Jieneng Chen at CCVL, on 3D spatial reasoning and AIGC. M.S., Biomedical Engineering — University of Science and Technology of China (USTC) Started Sept 2023 Advised by Prof. S. Kevin Zhou and Prof.https://mwxinnn.github.io/details/presentations/Mon, 01 Jan 0001 00:00:00 +0000https://mwxinnn.github.io/details/presentations/[Apr 2025] Invited talk, “Frontier Paper Sharing Session” — CSIG Wuhan Member Activity Center, Donghu Forumhttps://mwxinnn.github.io/details/teaching/Mon, 01 Jan 0001 00:00:00 +0000https://mwxinnn.github.io/details/teaching/ Teaching Assistant — Special Topic on Biomedical Engineering and Technological Innovation USTC · Started Sept 2024 Teaching Assistant — Frontiers of Electronic Information USTC · Sept 2023 – Jan 2024https://mwxinnn.github.io/details/honors/Mon, 01 Jan 0001 00:00:00 +0000https://mwxinnn.github.io/details/honors/[Nov 2024] 🎉 China National Scholarship by Chinese Ministry of Education (Top 0.2%) [Oct 2024] 🎉 Outstanding Student Scholarship (Grade 1) by USTC (Top 30%) [Oct 2023] 🎉 Outstanding Student Scholarship (Grade 1) by USTC (Top 30%) [Nov 2022] 🎉 Zhiyuan Honored Scholarship by SJTU (Top 5%) [Nov 2021] 🎉 Zhiyuan Honored Scholarship by SJTU (Top 5%) [Nov 2020] 🎉 Zhiyuan Honored Scholarship by SJTU (Top 5%) [Nov 2019] 🎉 Zhiyuan Honored Scholarship by SJTU (Top 5%)https://mwxinnn.github.io/details/news/Mon, 01 Jan 0001 00:00:00 +0000https://mwxinnn.github.io/details/news/[Mar 2026] Released Thinking with Spatial Code, exploring explicit 3D representations for video reasoning. [Jan 2026] Released CausalSpatial, a benchmark for object-centric causal spatial reasoning. [June 2025] 🎉 Two papers accepted at MICCAI 2025! [May 2025] 🎉 A paper accepted at ACL 2025 (Findings)! [Apr 2025] 🎤 Gave a talk at CSIG Wuhan Member Activity Center ‘Donghu Forum’ Frontier Paper Sharing Session! [Feb 2025] 🎉 A paper accepted at CVPR 2025!https://mwxinnn.github.io/tips/pytorch/bp/overview/Mon, 01 Jan 0001 00:00:00 +0000https://mwxinnn.github.io/tips/pytorch/bp/overview/BP Overview BP involves calculating the gradient of the loss function with respect to each weight by applying the chain rule, then updating the weights to minimize the loss. This process allows the model to learn from data. Computational Graph A computational graph is a representation of the operations and variables involved in a function or a network. It shows how each variable is computed from its predecessors and how they contribute to the final output.https://mwxinnn.github.io/details/publications/Mon, 01 Jan 0001 00:00:00 +0000https://mwxinnn.github.io/details/publications/Leading works 05 Participated works 11 Leading works First-author and co-first-author research. 2026 Spatial & video reasoningarXiv preprint · 2026 Spatial Code Help language models reason about the physical world by turning video into explicit 3D representations. Thinking with Spatial Code for Physical-World Video Reasoning Wenxin MaCo-first author Paper ↗Code ↗ Author listJieneng Chen†, Wenxin Ma†, Ruisheng Yuan†, Yunzhi Zhang†, Jiajun Wu, Alan Yuille Causal spatial reasoningarXiv preprint · 2026 CausalSpatial Test whether multimodal models can predict what happens when an object moves—across collision, compatibility, occlusion, and trajectory tasks.https://mwxinnn.github.io/details/research/Mon, 01 Jan 0001 00:00:00 +0000https://mwxinnn.github.io/details/research/I am interested in how AI agents actively explore their environments, gather useful information, and reason about how the world changes over time. My research focuses on: Active exploration: enabling agents to choose actions and observations that reduce uncertainty and build a better understanding of their environments. Spatial and temporal reasoning: learning representations that capture 3D structure, remain consistent over time, and support plausible cause-and-effect reasoning. Medical image analysis and anomaly detection: developing representations that help identify unusual patterns and support image segmentation.https://mwxinnn.github.io/tips/pytorch/transformers/customed_model/Mon, 01 Jan 0001 00:00:00 +0000https://mwxinnn.github.io/tips/pytorch/transformers/customed_model/