It’s possible that I shall make an ass of myself. But in that case one can always get out of it with a little dialectic. I have, of course, so worded my proposition as to be right either way (K.Marx, Letter to F.Engels on the Indian Mutiny)
Friday, January 23, 2026
BI-TE: Achieving GNN-based bandwidth indistinguishable traffic engineering in multi-domain SDN
The privacy protection issue of multi-domain traffic engineering has always been a hot topic. Consider that domains from different service providers may not be willing to expose the topology information within their own domains for security and privacy reasons. For example, in a cloud computing environment, different organizations may store data on the same cloud service platform. If the cloud service provider's security measures are insufficient, data leakage between different organizations may occur, and sensitive data of an organization may be accessed by other tenants, resulting in the leakage of business secrets or customer information. To solve the problems, a research team published their new research on 15 November 2025 in Frontiers of Computer Science co-published by Higher Education Press and Springer Nature.
The team adopt the concept of differential privacy and perturbed the domain information to achieve bandwidth indistinguishable TE. They first protected both the topology itself and the bandwidth information within the domain. Random perturbations were used for the topology. Then, Gaussian noise was added to the bandwidth information within the domain to achieve differential privacy protection. Unfortunately, perturbations may decrease the accuracy of the TE algorithm's resource allocation, negatively affecting performance. To mitigate this problem, they proposed BI-TE, which utilized a GNN-based bandwidth utilization prediction model to assist the controller in selecting the optimal forwarding path, thereby enhancing TE efficiency.
When a cross-domain request is sent to the root controller Root Controller (RC), RC inputs the perturbed network status information, cross-domain request, and routing information into the GNN model. If the predicted BU exceeds the threshold, this indicates that the bandwidth resources on the path may be insufficient. In this case, RC replaces the input routing path with the next shortest path, which is the second shortest path. Repeat this process until the prediction result of a certain path is lower than the threshold. Finally, transmission is performed based on the selected path.
Experimental results demonstrate that compared to abstraction-based hierarchical TE, BI-TE can reduce time overhead by nearly 24.35% while ensuring network bandwidth utilization close to 90%. Additionally, the fairness of allocation is also guaranteed.
Cognitive Diagnosis (CD) plays a crucial role in personalized learning by evaluating students' mastery of various concepts. However, current CD models face a significant challenge—the "student-concept sparsity barrier." This occurs when students have limited interactions with certain concepts.
To solve the problems, a research team led by Kun Zhang published their new research on 15 November 2025 in Frontiers of Computer Science co-published by Higher Education Press and Springer Nature.
The team proposed developed a new ESR-CD model. This model considers two key elements: concept mastery (how well students grasp concepts) and application ability (their capacity to apply knowledge in problem-solving). ESR-CD uses a unique sparsity-based mask module, along with matrix factorization and exercise-concept relationship refinement layers, enabling more accurate assessments even with limited data.
Extensive experiments demonstrated significant improvements of ESR-CD, with AUC gains of 1.5% on the ASSIST dataset and over 6% on the MOOC-Radar dataset. ESR-CD provides a more effective way to assess students' mastery and application skills in real-world learning environments. Accurate assessment can help educators create more personalized learning experiences. Future work can focus on exploring concept dependencies for better diagnostic performance by prompting large language models or preventing the amplification of unwanted biases in CD.
Researchers from Tianjin University have introduced the Emergency Medical Procedures 3D Dataset (EMP3D), a pioneering resource that captures the intricate movements of medical professionals during life-saving interventions with unprecedented precision. Published on 15 November 2025 in Frontiers of Computer Science co-published by Higher Education Press and Springer Nature, this dataset leverages synchronized multi-camera systems, advanced AI algorithms, and rigorous human validation to create the first 3D digital blueprint of emergency medical workflows. The innovation holds the potential to fundamentally transform emergency medical training and enhance robotic support in healthcare settings.
EMP3D in Action: A New Era for Emergency Care
The ultra-high precision of EMP3D can give rise to transformative downstream applications:
AI Coaches for Medics: The VR training platform can now evaluate and provide feedback on the trainee's performance in real time during operations such as chest compressions or hemostasis.
Rescue Robotics: Robots can mimic the actions of rescue workers and assist in carrying out rescue operations.
Crisis Analytics: Machine learning models trained on the EMP3D dataset can identify inefficiencies in a team's workflow during mass casualty incidents.
The Significance of EMP3D: Leverage metaverse technology to facilitate the widespread dissemination of emergency medical knowledge.
Current training tools for emergency medicine rely heavily on 2D videos or oversimplified simulations, which fail to capture the spatial complexity and split-second decisions required in real-life emergencies. This gap limits the effectiveness of AI-driven tools, robotic assistants, and virtual reality training platforms, which struggle to replicate the nuanced kinematics of human experts.
The EMP3D dataset directly addresses these challenges by offering:
1. High Precision Reconstruction: Unlike existing datasets, EMP3D captures 3D body motions, including fine finger movements (via SMPL-H models), essential for procedures such as fracture fixation and CPR.
2. AI-Ready Infrastructure: Every frame is manually validated, ensuring reliability for training machine learning models—a "gold standard" previously absent in emergency medicine.
3. Open Access: Freely available to researchers and developers, EMP3D accelerates innovation in healthcare AI and robotics.
From Multi-Camera Capture to Medical-Grade Models
The dataset’s creation involves a meticulously designed four-step pipeline:
1. Multi-view Chaos to Order: Six GoPro cameras, strategically positioned around an emergency room, capture synchronized video streams. We employ audio signals to achieve frame synchronization across multiple cameras. This alignment process effectively eliminates temporal drift, ensuring that the audio and video components remain synchronized throughout the recording.
2. Multi-view reconstruction: Using RTMPose algorithms, 2D poses are extracted from each camera view. Following this, the 4D association technique matches joints across perspectives, reconstructing 3D skeletal motion while handling occlusions and rapid movement.
3. Tracking in emergency medical settings: A custom Tracking Module maps the trajectories of rescuers and patients frame-by-frame, using feature vectors and cost-matrix optimization to resolve collisions in crowded scenarios.
4. Human-Perfected Modeling: Raw 3D joints are refined into SMPL-H body models via two-stage optimization. Every frame undergoes manual inspection.
Methodological workflow of EMP3D dataset generation
Researchers have developed a novel generative AI model, called Collaborative Competitive Agents (CCA), that significantly improves the ability to handle complex image editing tasks. This new approach utilizes multiple Large Language Model (LLM)-based agents that work both collaboratively and competitively, resulting in a more robust and accurate editing process compared to existing methods. This breakthrough allows for a more transparent and iterative approach to image manipulation, enabling a level of precision previously unattainable. The findings were published on 15 November 2025 in Frontiers of Computer Science, co-published by Higher Education Press and Springer Nature.
The CCA system, developed by a team led by Tiankai Hang, Shuyang Gu, Dong Chen, Xin Geng, and Baining Guo from Southeast University and Microsoft Research Asia, draws inspiration from Generative Adversarial Networks (GANs). Unlike traditional "black box" generative models, CCA allows for observation and control of intermediate steps. It employs two "generator" agents that independently process user instructions and create edited images, and a "discriminator" agent that evaluates the results and provides feedback. This feedback loop, combined with the competitive dynamic between the generator agents, leads to continuous improvement and refinement of the output.
"Existing image editing tools often struggle with complex, multi-step instructions," explains Tiankai Hang, the first author of the study. "Our CCA system leverages the power of LLMs to decompose these complex tasks into manageable sub-tasks, and the collaborative-competitive nature of the agents ensures that the final result closely matches the user's intent."
The key innovation of CCA lies in its multi-agent architecture and the relationships between these agents. The generator agents not only learn from the discriminator's feedback but also from each other's successes and failures. This transparency and iterative optimization process are crucial for handling intricate editing requests, such as "colorizing an old photograph, replacing the depicted individual with the user's image, and adding a hoe in the user's hand." Such complex commands often stump conventional image editing software.
The research team demonstrated the effectiveness of CCA through comprehensive experiments, comparing it to several state-of-the-art image editing techniques. The results showed that CCA consistently outperformed these methods, particularly when dealing with complex instructions. Human preference studies also indicated that users found CCA's outputs to be more aligned with their requirements and of higher overall quality.
While the current study focuses on image editing, the CCA framework is versatile and has the potential to be applied to other generative tasks, such as text-to-image generation. The researchers envision further applications in areas requiring complex reasoning and analysis, highlighting the broader impact of this work beyond the creative industries.
