Inside Alibaba AI Lab: Dr. Wang Gang on Multimodal AI and Edge Computing

Interview Overview: A Single Model for All Problems Is Unrealistic

Dr. Wang Gang, a distinguished scientist at Alibaba AI Lab, argues that Google’s claim of "one model solves all problems" is unrealistic. He emphasizes the need for domain‑specific model customization and highlights the lab’s focus on multimodal training, combining voice and vision for future human‑machine interaction.

One Person Secures Three CVPR Papers

At CVPR 2017, Alibaba AI Lab presented three papers, each with Dr. Wang’s deep involvement:

Deep Level Sets for Salient Object Detection – a method that merges deep networks with level‑set techniques to improve segmentation detail.

Global Context‑Aware Attention LSTM Networks for 3D Action Recognition – introduces a context‑aware attention mechanism within LSTM to achieve state‑of‑the‑art results on multiple 3D action datasets.

Episodic CAMN: Contextual Attention‑based Memory Networks With Iterative Feedback For Scene Labeling – applies iterative feedback attention to scene segmentation, useful for autonomous driving.

These works address salient object detection, action recognition, and scene labeling, and illustrate the lab’s commitment to translating deep‑learning research into practical applications such as home‑security monitoring.

Voice + Vision Is the Future of Interaction

Dr. Wang believes that combining speech and visual cues is essential for next‑generation interaction. He notes that Alibaba’s smart speaker (Tmall Genie X1) integrates these modalities, and that multimodal research is a hot topic in both academia and industry.

From Recognition to Understanding, a Fundamental Link Is Missing

The scientist points out that current AI excels at recognition but lacks true understanding because it misses world knowledge and reasoning. He stresses that defining clear tasks is a prerequisite for research, and that industrial applications often drive the need for deeper understanding.

Research Interests: Edge AI, Bio‑Inspired Cognition, Quantum Computing

Dr. Wang’s personal research interests include developing lightweight neural networks for edge devices, drawing inspiration from biological cognition to design more efficient architectures, and exploring the synergy between quantum computing and deep learning to accelerate training.

Consumer AI Products: Personal Interest and Lab Focus

Joining Alibaba allowed Dr. Wang to work on consumer‑grade AI products close to end users. He finds C‑side development “sexier” than cloud‑only solutions and sees great potential in bringing AI capabilities into everyday devices.

Alibaba's AI Talent Program

Starting summer 2017, Alibaba launched a campus recruitment program targeting Ph.D. candidates in computer vision, machine learning, NLP, graphics, and speech interaction. The lab also announced the “NASA” initiative to build a long‑term independent R&D department covering AI, chips, IoT, operating systems, and biometric technologies.