Hybrid Online-Offline Learning for Task Offloading in Mobile Edge Computing Systems

We consider a multi-user multi-server mobile edge computing (MEC) system, in which users arrive on a network randomly over time and generate computation tasks, which will be computed either locally on their own computing devices or be offloaded to one of the MEC servers. Under such a dynamic network environment, we propose a novel task offloading policy based on hybrid online-offline learning, which can efficiently reduce the overall computation delay and energy consumption only with information available at nearest MEC servers from each user. We provide a practical signaling and learning framework that can train deep neural networks for both online and offline learning and can adjust its offloading policy based on the queuing status of each MEC server and network dynamics. Numerical results demonstrate that the proposed scheme significantly reduces the average computation delay for a broad class of network environments compared to the conventional offloading methods. It is further shown that the proposed hybrid online-offline learning framework can be extended to a general cost function reflecting both delay and energy-dependent metrics.

Scaling physics-informed hard constraints with mixture-of-experts

Imposing known physical constraints, such as conservation laws, during neural network training introduces an inductive bias that can improve accuracy, reliability, convergence, and data efficiency for modeling physical dynamics. While such constraints can be softly imposed via loss function penalties, recent advancements in differentiable physics and optimization improve performance by incorporating PDE-constrained optimization as individual layers in neural networks. This enables a stricter adherence to physical constraints. However, imposing hard constraints significantly increases computational and memory costs, especially for complex dynamical systems. This is because it requires solving an optimization problem over a large number of points in a mesh, representing spatial and temporal discretizations, which greatly increases the complexity of the constraint. To address this challenge, we develop a scalable approach to enforce hard physical constraints using Mixture-of-Experts (MoE), which can be used with any neural network architecture. Our approach imposes the constraint over smaller decomposed domains, each of which is solved by an "expert" through differentiable optimization. During training, each expert independently performs a localized backpropagation step by leveraging the implicit function theorem; the independence of each expert allows for parallelization across multiple GPUs. Compared to standard differentiable optimization, our scalable approach achieves greater accuracy in the neural PDE solver setting for predicting the dynamics of challenging non-linear systems. We also improve training stability and require significantly less computation time during both training and inference stages.

FIDLAR: Forecast-Informed Deep Learning Architecture for Flood Mitigation

In coastal river systems, frequent floods, often occurring during major storms or king tides, pose a severe threat to lives and property. However, these floods can be mitigated or even prevented by strategically releasing water before extreme weather events with hydraulic structures such as dams, gates, pumps, and reservoirs. A standard approach used by local water management agencies is the "rule-based" method, which specifies predetermined pre-releases of water based on historical and time-tested human experience, but which tends to result in excess or inadequate water release. The model predictive control (MPC), a physics-based model for prediction, is an alternative approach, albeit involving computationally intensive calculations. In this paper, we propose a Forecast Informed Deep Learning Architecture, FIDLAR, to achieve rapid and optimal flood management with precise water pre-releases. FIDLAR seamlessly integrates two neural network modules: one called the Flood Manager, which is responsible for generating water pre-release schedules, and another called the Flood Evaluator, which assesses these generated schedules. The Evaluator module is pre-trained separately, and its gradient-based feedback is used to train the Manager model, ensuring optimal water pre-releases. We have conducted experiments using FIDLAR with data from a flood-prone coastal area in South Florida, particularly susceptible to frequent storms. Results show that FIDLAR is several orders of magnitude faster than currently used physics-based approaches while outperforming baseline methods with improved water pre-release schedules. Our code is at https://github.com/JimengShi/FIDLAR/.

Robust Learning-Augmented Dictionaries

We present the first learning-augmented data structure for implementing dictionaries with optimal consistency and robustness. Our data structure, named RobustSL, is a skip list augmented by predictions of access frequencies of elements in a data sequence. With proper predictions, RobustSL has optimal consistency (achieves static optimality). At the same time, it maintains a logarithmic running time for each operation, ensuring optimal robustness, even if predictions are generated adversarially. Therefore, RobustSL has all the advantages of the recent learning-augmented data structures of Lin, Luo, and Woodruff (ICML 2022) and Cao et al. (arXiv 2023), while providing robustness guarantees that are absent in the previous work. Numerical experiments show that RobustSL outperforms alternative data structures using both synthetic and real datasets.

SusFL: Energy-Aware Federated Learning-based Monitoring for Sustainable Smart Farms

We propose a novel energy-aware federated learning (FL)-based system, namely SusFL, for sustainable smart farming to address the challenge of inconsistent health monitoring due to fluctuating energy levels of solar sensors. This system equips animals, such as cattle, with solar sensors with computational capabilities, including Raspberry Pis, to train a local deep-learning model on health data. These sensors periodically update Long Range (LoRa) gateways, forming a wireless sensor network (WSN) to detect diseases like mastitis. Our proposed SusFL system incorporates mechanism design, a game theory concept, for intelligent client selection to optimize monitoring quality while minimizing energy use. This strategy ensures the system's sustainability and resilience against adversarial attacks, including data poisoning and privacy threats, that could disrupt FL operations. Through extensive comparative analysis using real-time datasets, we demonstrate that our FL-based monitoring system significantly outperforms existing methods in prediction accuracy, operational efficiency, system reliability (i.e., mean time between failures or MTBF), and social welfare maximization by the mechanism designer. Our findings validate the superiority of our system for effective and sustainable animal health monitoring in smart farms. The experimental results show that SusFL significantly improves system performance, including a $10\%$ reduction in energy consumption, a $15\%$ increase in social welfare, and a $34\%$ rise in Mean Time Between Failures (MTBF), alongside a marginal increase in the global model's prediction accuracy.

A Recurrent Neural Network Enhanced Unscented Kalman Filter for Human Motion Prediction

This paper presents a deep learning enhanced adaptive unscented Kalman filter (UKF) for predicting human arm motion in the context of manufacturing. Unlike previous network-based methods that solely rely on captured human motion data, which is represented as bone vectors in this paper, we incorporate a human arm dynamic model into the motion prediction algorithm and use the UKF to iteratively forecast human arm motions. Specifically, a Lagrangian-mechanics-based physical model is employed to correlate arm motions with associated muscle forces. Then a Recurrent Neural Network (RNN) is integrated into the framework to predict future muscle forces, which are transferred back to future arm motions based on the dynamic model. Given the absence of measurement data for future human motions that can be input into the UKF to update the state, we integrate another RNN to directly predict human future motions and treat the prediction as surrogate measurement data fed into the UKF. A noteworthy aspect of this study involves the quantification of uncertainties associated with both the data-driven and physical models in one unified framework. These quantified uncertainties are used to dynamically adapt the measurement and process noises of the UKF over time. This adaption, driven by the uncertainties of the RNN models, addresses inaccuracies stemming from the data-driven model and mitigates discrepancies between the assumed and true physical models, ultimately enhancing the accuracy and robustness of our predictions. Compared to the traditional RNN-based prediction, our method demonstrates improved accuracy and robustness in extensive experimental validations of various types of human motions.

How Temporal Unrolling Supports Neural Physics Simulators

Unrolling training trajectories over time strongly influences the inference accuracy of neural network-augmented physics simulators. We analyze these effects by studying three variants of training neural networks on discrete ground truth trajectories. In addition to commonly used one-step setups and fully differentiable unrolling, we include a third, less widely used variant: unrolling without temporal gradients. Comparing networks trained with these three modalities makes it possible to disentangle the two dominant effects of unrolling, training distribution shift and long-term gradients. We present a detailed study across physical systems, network sizes, network architectures, training setups, and test scenarios. It provides an empirical basis for our main findings: A non-differentiable but unrolled training setup supported by a numerical solver can yield 4.5-fold improvements over a fully differentiable prediction setup that does not utilize this solver. We also quantify a difference in the accuracy of models trained in a fully differentiable setup compared to their non-differentiable counterparts. While differentiable setups perform best, the accuracy of unrolling without temporal gradients comes comparatively close. Furthermore, we empirically show that these behaviors are invariant to changes in the underlying physical system, the network architecture and size, and the numerical scheme. These results motivate integrating non-differentiable numerical simulators into training setups even if full differentiability is unavailable. We also observe that the convergence rate of common neural architectures is low compared to numerical algorithms. This encourages the use of hybrid approaches combining neural and numerical algorithms to utilize the benefits of both.

MapTrack: Tracking in the Map

Multi-Object Tracking (MOT) aims to maintain stable and uninterrupted trajectories for each target. Most state-of-the-art approaches first detect objects in each frame and then implement data association between new detections and existing tracks using motion models and appearance similarities. Despite achieving satisfactory results, occlusion and crowds can easily lead to missing and distorted detections, followed by missing and false associations. In this paper, we first revisit the classic tracker DeepSORT, enhancing its robustness over crowds and occlusion significantly by placing greater trust in predictions when detections are unavailable or of low quality in crowded and occluded scenes. Specifically, we propose a new framework comprising of three lightweight and plug-and-play algorithms: the probability map, the prediction map, and the covariance adaptive Kalman filter. The probability map identifies whether undetected objects have genuinely disappeared from view (e.g., out of the image or entered a building) or are only temporarily undetected due to occlusion or other reasons. Trajectories of undetected targets that are still within the probability map are extended by state estimations directly. The prediction map determines whether an object is in a crowd, and we prioritize state estimations over observations when severe deformation of observations occurs, accomplished through the covariance adaptive Kalman filter. The proposed method, named MapTrack, achieves state-of-the-art results on popular multi-object tracking benchmarks such as MOT17 and MOT20. Despite its superior performance, our method remains simple, online, and real-time. The code will be open-sourced later.

Critic-Actor for Average Reward MDPs with Function Approximation: A Finite-Time Analysis

In recent years, there has been a lot of research work activity focused on carrying out asymptotic and non-asymptotic convergence analyses for two-timescale actor critic algorithms where the actor updates are performed on a timescale that is slower than that of the critic. In a recent work, the critic-actor algorithm has been presented for the infinite horizon discounted cost setting in the look-up table case where the timescales of the actor and the critic are reversed and asymptotic convergence analysis has been presented. In our work, we present the first critic-actor algorithm with function approximation and in the long-run average reward setting and present the first finite-time (non-asymptotic) analysis of such a scheme. We obtain optimal learning rates and prove that our algorithm achieves a sample complexity of $\mathcal{\tilde{O}}(\epsilon^{-2.08})$ for the mean squared error of the critic to be upper bounded by $\epsilon$ which is better than the one obtained for actor-critic in a similar setting. We also show the results of numerical experiments on three benchmark settings and observe that the critic-actor algorithm competes well with the actor-critic algorithm.

A Critical Evaluation of AI Feedback for Aligning Large Language Models

Reinforcement learning with AI feedback (RLAIF) is a popular paradigm for improving the instruction-following abilities of powerful pre-trained language models. RLAIF first performs supervised fine-tuning (SFT) using demonstrations from a teacher model and then further fine-tunes the model with reinforcement learning (RL), using feedback from a critic model. While recent popular open-source models have demonstrated substantial improvements in performance from the RL step, in this paper we question whether the complexity of this RL step is truly warranted for AI feedback. We show that the improvements of the RL step are virtually entirely due to the widespread practice of using a weaker teacher model (e.g. GPT-3.5) for SFT data collection than the critic (e.g., GPT-4) used for AI feedback generation. Specifically, we show that simple supervised fine-tuning with GPT-4 as the teacher outperforms existing RLAIF pipelines. More generally, we find that the gains from RLAIF vary substantially across base model families, test-time evaluation protocols, and critic models. Finally, we provide a mechanistic explanation for when SFT may outperform the full two-step RLAIF pipeline as well as suggestions for making RLAIF maximally useful in practice.