🏢 Shanghai Jiao Tong University

Enhance deep neural network privacy and trustworthiness with unified gradient-based machine unlearning, leveraging remain geometry for efficient forgetting and performance preservation.

DNNs learn interactions in two phases: initially removing complex interactions, then gradually learning higher-order ones, leading to overfitting.

Softmax regression reveals in-context learning’s surprising similarity to gradient descent in self-attention Transformers, showing the models’ remarkable learning capabilities.

SFCNet, a novel spherical frustum sparse convolution network, tackles LiDAR point cloud semantic segmentation by eliminating quantized information loss, leading to superior performance, especially for…

Self-Calibrated Tuning (SCT) enhances vision-language model OOD detection by adaptively weighting OOD regularization based on prediction uncertainty, mitigating issues caused by inaccurate feature ext…

SeeA* enhances A* search by selectively sampling promising nodes, improving exploration and efficiency, especially with less accurate heuristics.

SceneCraft generates highly detailed indoor scenes from user-provided textual descriptions and spatial layouts, overcoming limitations of previous text-to-3D methods in scale and control.

DisCo: a novel framework for generalizable complex image generation using scene graph disentanglement and composition, achieving superior performance over existing methods.

Enhanced VQAs via Hamming Weight Preserving ansatz and parity checks achieve superior performance on quantum chemistry and combinatorial problems, showcasing quantum advantage potential in NISQ era.

ResAD, a novel framework, tackles class-generalizable anomaly detection by learning residual feature distributions, achieving remarkable results on diverse datasets without retraining.

ReLIZO boosts zeroth-order optimization by cleverly reusing past queries, drastically cutting computation costs while maintaining gradient estimation accuracy.

POAD enhances LLM agents by decomposing language agent optimization to the token level, achieving finer-grained credit assignment and improved learning efficiency and generalization.

Quantum VAE with spherical latent variable learning enables efficient, one-shot 3D molecule generation, outperforming classic and other quantum methods.

QuadMamba: A novel vision model leveraging quadtree-based scanning for superior performance in visual tasks, achieving state-of-the-art results with linear-time complexity.

Enhance multimodal model robustness against missing data with Probabilistic Conformal Distillation (PCD)! PCD models missing modalities probabilistically, achieving superior performance on multiple be…

PCP-MAE enhances point cloud self-supervised learning by cleverly predicting masked patch centers, leading to superior 3D object classification and scene segmentation.

MM-Det, a novel algorithm, uses multimodal learning and spatiotemporal attention to detect diffusion-generated videos, achieving state-of-the-art performance on the new DVF dataset.

Nimbus achieves 2.7-4.7x speedup in BERT base inference using novel two-party computation techniques for efficient matrix multiplication and non-linear layer approximation.

NeuRodin: A two-stage neural framework achieves high-fidelity 3D surface reconstruction from posed RGB images by innovatively addressing limitations in SDF-based methods, resulting in superior reconst…

Ref-MC2 reconstructs high-fidelity 3D objects with inter-reflections by using a novel multi-times Monte Carlo sampling strategy, achieving superior performance in accuracy and efficiency.