Weekly Developer Highlights: Flutter 2, JD Cloud, Flink 1.12.2, AI Breakthroughs

Flutter 2 Released

On March 4, Google announced Flutter 2, a portable framework that expands Flutter from a mobile‑only SDK to support web, desktop, and embedded devices, enabling developers to use a single codebase for iOS, Android, Windows, macOS, Linux, browsers, cars, TVs, and appliances.

JD Cloud’s Next‑Gen Self‑Developed Server

JD Cloud will launch a new self‑developed cloud server in April 2021, offering more than double the instance capacity at the same price, supporting internal JD applications and external services, built on a self‑designed architecture and planned to be open‑sourced to the industry.

Apache Flink 1.12.2 Released

Apache Flink 1.12.2 focuses on bug fixes, including a memory‑management issue where hosted memory was not released promptly, correction of a typo in the EXPLAIN statement documentation, and updates on binary compatibility with Scala after version 2.12.8.

Sidewalk Robots Recognized as Pedestrians

In Pennsylvania and other states, sidewalk robots are legally classified as pedestrians; Starship reports that about 1,000 such robots are operating publicly, highlighting a regulatory shift that treats them like ordinary foot traffic.

Microsoft Mesh Mixed‑Reality Platform

Microsoft unveiled Mesh, a mixed‑reality cloud platform that delivers immersive, real‑time collaboration across devices such as phones, tablets, PCs, and VR headsets, with applications including a HoloLens preview, AltspaceVR integration, and future extensions to Teams and Dynamics 365.

Facebook’s SEER Self‑Supervised AI Model

Facebook introduced SEER (Self‑Supervised), a vision model trained on one billion unlabelled Instagram images, capable of learning without manual annotations and achieving state‑of‑the‑art accuracy on object detection tasks.

EMNLP 2020 – Text‑Enhanced Knowledge‑Graph QA

The paper proposes a QA method that enriches knowledge‑graph entities with semantic information extracted from text, treating text as hyper‑edges and applying hyper‑graph convolutional networks, demonstrating improved performance on the WebQuestionSP benchmark.

COLING 2020 – Bidirectional Cognitive Thinking Network

Inspired by human cognition, the authors design a neural network that simulates reverse and inertia thinking to enhance machine reading comprehension, achieving better results by integrating these cognitive patterns.