More Like this

1. Differentiable Weightless Neural Networks

   Bacellar, Alan; Susskind, Zachary; Breternitz, Maurício; John, Eugene; John, Lizy; Lima, Priscila; Franca, Felipe (July 2024, Proceedings of Machine Learning Research)

   We introduce the Differentiable Weightless Neural Network (DWN), a model based on interconnected lookup tables. Training of DWNs is enabled by a novel Extended Finite Difference technique for approximate differentiation of binary values. We propose Learnable Mapping, Learnable Reduction, and Spectral Regularization to further improve the accuracy and efficiency of these models. We evaluate DWNs in three edge computing contexts: (1) an FPGA-based hardware accelerator, where they demonstrate superior latency, throughput, energy efficiency, and model area compared to state-of-the-art solutions, (2) a low-power microcontroller, where they achieve preferable accuracy to XGBoost while subject to stringent memory constraints, and (3) ultra-low-cost chips, where they consistently outperform small models in both accuracy and projected hardware area. DWNs also compare favorably against leading approaches for tabular datasets, with higher average rank. Overall, our work positions DWNs as a pioneering solution for edge-compatible high-throughput neural networks. 

   more » « less
2. DepS: Delayed epsilon-Shrinking for Faster Once-for-All Training

   https://doi.org/10.1007/978-3-031-73024-5_19  

   Annavajjala, Aditya; Khare, Alind; Agrawal, Animesh; Fedorov, Igor; Latapie, Hugo; Lee, Myungjin; Tumanov, Alexey (November 2024, Springer Nature Switzerland)

   CNNs are increasingly deployed across different hardware, dynamic environments, and low-power embedded devices. This has led to the design and training of CNN architectures with the goal of maximizing accuracy subject to such variable deployment constraints. As the number of deployment scenarios grows, there is a need to find scalable solutions to design and train specialized CNNs. Once-for-all training has emerged as a scalable approach that jointly co-trains many models (subnets) at once with a constant training cost and finds specialized CNNs later. The scalability is achieved by training the full model and simultaneously reducing it to smaller subnets that share model weights (weight-shared shrinking). However, existing once-for-all training approaches incur huge training costs reaching 1200 GPU hours. We argue this is because they either start the process of shrinking the full model too early or too late. Hence, we propose Delayed Epsilon-Shrinking (DepS) that starts the process of shrinking the full model when it is partially trained, which leads to training cost improvement and better in-place knowledge distillation to smaller models. The proposed approach also consists of novel heuristics that dynamically adjust subnet learning rates incrementally, leading to improved weight-shared knowledge distillation from larger to smaller subnets as well. As a result, DepS outperforms state-of-the-art once-for-all training techniques across different datasets including CIFAR10/100, ImageNet-100, and ImageNet-1k on accuracy and cost. It achieves higher ImageNet-1k top1 accuracy or the same accuracy with 1.3x reduction in FLOPs and 2.5x drop in training cost (GPU*hrs). 

   more » « less
3. LLMCO2: Advancing Accurate Carbon Footprint Prediction for LLM Inferences

   https://doi.org/10.1145/3757892.3757901  

   Fu, Zhenxiao; Chen, Fan; Zhou, Shan; Li, Haitong; Jiang, Lei (July 2025, ACM SIGEnergy Energy Informatics Review)

   Throughout its lifecycle, an LLM incurs significantly higher carbon emissions during inference than training. Inference requests vary in batch size, prompt length, and token generation, while cloud providers deploy heterogeneous GPU configurations to meet diverse service-level objectives. Unlike training, inference exhibits lower and highly variable hardware utilization, making equation-based carbon models unreliable. Existing network-based estimators lack accuracy, as they fail to account for the distinct prefill and decode phases, hardware-specific features, and realistic request distributions. We propose LLMCO₂, a graph neural network (GNN)-based model, to improve the accuracy of LLM inference carbon footprint estimation by ~ 67% over prior approaches. Source code is available at https://github.com/fuzhenxiao/LLMCO₂. 

   more » « less
4. Fed-CASQ: Enhancing Class-Wise Accuracy in Pervasive Federated Learning with Class-Aware Scaling and Quantization

   https://doi.org/10.1109/PerCom67906.2026.11524552  

   Dey, Emon; Ravi, Anuradha; Shinde, Gaurav; Chugh, Garvit; Ghosh, Indrajeet; Misra, Archan; Roy, Nirmalya (March 2026, IEEE)

   Federated Learning (FL) enables collaborative machine learning across decentralized devices and data sources, but resource constraints on pervasive devices necessitate efficient model compression. Existing approaches, such as quantization for on-device training, often degrade accuracy, especially for classes that are difficult to learn due to imbalance, poor-quality samples, or inherent complexity. This results in persistent accuracy gaps across classes. We propose Fed-CASQ's a novel framework that couples class-aware strategies into the quantization process to jointly improve efficiency and accuracy in pervasive FL. Unlike prior works that address quantization and imbalance separately, Fed-CASQ adaptively selects quantization levels based on device resources and leverages Layer-wise Relevance Propagation (LRP) to assess class-relevant convolutional neural network (CNN) filters on the client side. An adaptive weight scaling mechanism is then applied to amplify critical information for low-accuracy classes before aggregation. At the server, a complementary novel aggregation strategy mitigates global imbalance across clients, ensuring that underperforming classes receive proportional attention during model updates. We theoretically establish that Fed-CASQ achieves a convergence rate of O(κ∗σ^∗δ^T√) under non-convex settings. We empirically establish that quantization directly influences the performance of under sampled (minority) classes. Experimental results further show that Fed-CASQ substantially narrows the performance gap for low-accuracy classes, improving their accuracy by ≈30%, while reducing training latency by over 56% on resource-constrained pervasive devices. 

   more » « less
5. SPLIDT: Partitioned Decision Trees for Scalable Stateful Inference at Line Rate

   Parvez, Murayyiam; Zulfiqar, Annus; Beltiukov, Roman; Landau_Feibish, Shir; Willinger, Walter; Gupta, Arpit; Shahbaz, Muhammad (May 2026, 23rd USENIX Symposium on Networked Systems Design and Implementation (NSDI '26))

   Machine learning (ML) is increasingly being deployed in programmable data planes (switches and SmartNICs) to enable real-time traffic analysis, security monitoring, and in-network decision-making. Decision trees (DTs) are particularly well-suited for these tasks due to their interpretability and compatibility with data-plane architectures, i.e., match-action tables (MATs). However, existing in-network DT implementations are constrained by the need to compute all input features upfront, forcing models to rely on a small, fixed set of features per flow. This significantly limits model accuracy and scalability under stringent hardware resource constraints. We present SPLIDT, a system that rethinks DT deployment in the data plane by enabling partitioned inference over sliding windows of packets. SPLIDT introduces two key innovations: (1) it groups individual subtrees of a DT into partitions and allows each subtree to have its own feature set, and (2) it leverages an in-band control channel (via recirculation) to reuse data-plane resources (both stateful registers and match keys) across partitions at line rate. These insights allow SPLIDT to scale the number of stateful features a model can use without exceeding hardware limits. To support this architecture, SPLIDT incorporates a custom training and design-space exploration (DSE) framework that jointly optimizes feature allocation, tree partitioning, and DT model depth. Evaluation across multiple real-world datasets shows that SPLIDT achieves higher accuracy while supporting up to 5x more stateful features than prior approaches (e.g., NetBeacon and Leo). It maintains the same low time-to-detection (TTD) as these systems, while scaling to millions of flows with minimal recirculation overhead (≤ 0.05%). 

   more » « less