Deploying DNNs in P4 switches seemed impossible—until we distilled them into cascaded LUTs. The result? Lossless accuracy, wirespeed performance, and 96% less memory than traditional methods. In-network AI just got real. 🌟 Follow us on LinkedIn! Check the updates…
No problem. Our cascaded LUT distillation maps DNNs into P4 flow tables, enabling wirespeed inference without hardware changes. Tested on DDoS detection, it achieves 93% F1-score with 8-bit inputs. 🌟 Follow us on LinkedIn! Check the updates from the…
Edge AI workloads demand low-latency, privacy-compliant storage. CATER delivers, using policy-driven placement to optimize energy, latency, and compliance. Tested on real edge clusters, it cuts data movement by 61%. 🌟 Follow us on LinkedIn! Check the updates from the…
Combining LSTM for QoS prediction and LOF for anomaly detection, our predictive retraining adapts to real-time traffic changes in B5G networks. The result? Fewer violations, smarter resource use, and 99% accuracy in detecting new patterns. 🌟 Follow us on LinkedIn!…
Static retraining fails in dynamic 6G networks. Our predictive approach uses LOF and LSTM to detect traffic shifts and retrain just in time, reducing SLA violations by 40% compared to threshold-based methods. 🌟 Follow us on LinkedIn! Check the…
Apache Ozone excels at scalable storage—but lacks constraint-aware placement. CATER fills the gap, integrating seamlessly to reduce active nodes by 23% while respecting GDPR, hardware, and collocation rules. 🌟 Follow us on LinkedIn! Check the updates from the website:…
RDMA’s memory registration can kill performance—unless you pre-register buffers. Our GVirtuS RDMA communicator does exactly that, slashing context switches by 98% and boosting GPGPU virtualization speed by up to 55%. 🌟 Follow us on LinkedIn! Check the updates from…
Virtualizing GPUs over TCP/IP introduces latency and context switches. Our RDMA communicator for GVirtuS cuts Matrix Multiplication time by 5.5x and SAXPY by 82%, proving Infiniband is the future of low-latency GPGPU remoting. 🌟 Follow us on LinkedIn! Check…
P4 switches weren’t built for DNNs—but our LUT distillation changes that. By mapping neurons to flow tables, we enable complex AI models in hardware-constrained pipelines, with no accuracy loss. Tested on DDoS detection, it’s faster and leaner than GPU offloading.…
Detecting DDoS attacks at wirespeed requires in-network AI. Our P4 LUT distillation deploys a DNN-based mitigator without ALUs, achieving 93% F1-score on UNSW-NB15 dataset. The secret? Cascaded lookup tables that replace neurons with flow matches. 🌟 Follow us on LinkedIn!…
