Ant Group’s Six AI Papers Presented at AAAI 2020

On February 7, 2020, the 34th AAAI Conference (AAAI 2020) was held in New York, and Ant Group had six papers accepted, reflecting its strong presence in the top AI venue.

Paper 1 – Dynamic Network Pruning with Interpretable Layerwise Channel Selection : Proposes a dynamic pruning method that selects channels discretely, enabling real‑time inference paths, visualizable decisions, and improved robustness against adversarial attacks.

Paper 2 – Pruning from Scratch : Demonstrates that effective pruning structures can be obtained directly from randomly initialized weights, eliminating the costly pre‑training stage and yielding more diverse, higher‑performing models.

Paper 3 – Automatic Car‑Insurance Damage Assessment System: Reading and Understanding Video Like a Professional Adjuster : Introduces an AI‑driven system that processes user‑captured videos in the cloud to produce accurate damage estimates within seconds, reducing both time and labor costs.

Paper 4 – Long Short‑Term Sample Distillation (LSTSD) : Leverages long‑term and short‑term historical training signals as teachers for a student model, allowing simultaneous learning from multiple past states and achieving superior performance on NLP and CV tasks.

Paper 5 – Span‑based Neural Buffer: Towards Efficient and Effective Utilization of Long‑Distance Context for Neural Sequence Models : Presents a block‑wise tensor buffer that stores contextual chunks, enabling models to capture ultra‑long dependencies efficiently, with adaptive reward‑based training to mitigate sample bias.

Paper 6 – Cost‑Effective Incentive Allocation via Structured Counterfactual Inference : Formulates incentive distribution as a counterfactual feedback policy optimization problem, converting it to a domain‑adaptation task and providing theoretical error bounds and empirical gains under budget constraints.

Overall, these works illustrate Ant Group’s commitment to advancing AI research across model compression, computer vision, natural language processing, and data‑driven decision making.