Neural Bandit Based Optimal LLM Selection for a Pipeline of Tasks

With the increasing popularity of large language models (LLMs) for a variety of tasks, there has been a growing interest in strategies that can predict which out of a set of LLMs will yield a successful answer at low cost. This problem promises to become more and more relevant as providers like Microsoft allow users to easily create custom LLM "assistants" specialized to particular types of queries. However, some tasks (i.e., queries) may be too specialized and difficult for a single LLM to handle alone. These applications often benefit from breaking down the task into smaller subtasks, each of which can then be executed by a LLM expected to perform well on that specific subtask. For example, in extracting a diagnosis from medical records, one can first select an LLM to summarize the record, select another to validate the summary, and then select another, possibly different, LLM to extract the diagnosis from the summarized record. Unlike existing LLM selection or routing algorithms, this setting requires that we select a sequence of LLMs, with the output of each LLM feeding into the next and potentially influencing its success. Thus, unlike single LLM selection, the quality of each subtask's output directly affects the inputs, and hence the cost and success rate, of downstream LLMs, creating complex performance dependencies that must be learned and accounted for during selection. We propose a neural contextual bandit-based algorithm that trains neural networks that model LLM success on each subtask in an online manner, thus learning to guide the LLM selections for the different subtasks, even in the absence of historical LLM performance data. Experiments on telecommunications question answering and medical diagnosis prediction datasets illustrate the effectiveness of our proposed approach compared to other LLM selection algorithms.

Semantic Caching for Low-Cost LLM Serving: From Offline Learning to Online Adaptation

Large Language Models (LLMs) are revolutionizing how users interact with information systems, yet their high inference cost poses serious scalability and sustainability challenges. Caching inference responses, allowing them to be retrieved without another forward pass through the LLM, has emerged as one possible solution. Traditional exact-match caching, however, overlooks the semantic similarity between queries, leading to unnecessary recomputation. Semantic caching addresses this by retrieving responses based on semantic similarity, but introduces a fundamentally different cache eviction problem: one must account for mismatch costs between incoming queries and cached responses. Moreover, key system parameters, such as query arrival probabilities and serving costs, are often unknown and must be learned over time. Existing semantic caching methods are largely ad-hoc, lacking theoretical foundations and unable to adapt to real-world uncertainty. In this paper, we present a principled, learning-based framework for semantic cache eviction under unknown query and cost distributions. We formulate both offline optimization and online learning variants of the problem, and develop provably efficient algorithms with state-of-the-art guarantees. We also evaluate our framework on a synthetic dataset, showing that our proposed algorithms perform matching or superior performance compared with baselines.

Evaluating Selective Encryption Against Gradient Inversion Attacks

Gradient inversion attacks pose significant privacy threats to distributed training frameworks such as federated learning, enabling malicious parties to reconstruct sensitive local training data from gradient communications between clients and an aggregation server during the aggregation process. While traditional encryption-based defenses, such as homomorphic encryption, offer strong privacy guarantees without compromising model utility, they often incur prohibitive computational overheads. To mitigate this, selective encryption has emerged as a promising approach, encrypting only a subset of gradient data based on the data's significance under a certain metric. However, there have been few systematic studies on how to specify this metric in practice. This paper systematically evaluates selective encryption methods with different significance metrics against state-of-the-art attacks. Our findings demonstrate the feasibility of selective encryption in reducing computational overhead while maintaining resilience against attacks. We propose a distance-based significance analysis framework that provides theoretical foundations for selecting critical gradient elements for encryption. Through extensive experiments on different model architectures (LeNet, CNN, BERT, GPT-2) and attack types, we identify gradient magnitude as a generally effective metric for protection against optimization-based gradient inversions. However, we also observe that no single selective encryption strategy is universally optimal across all attack scenarios, and we provide guidelines for choosing appropriate strategies for different model architectures and privacy requirements.

Offline Clustering of Linear Bandits: Unlocking the Power of Clusters in Data-Limited Environments

Contextual linear multi-armed bandits are a learning framework for making a sequence of decisions, e.g., advertising recommendations for a sequence of arriving users. Recent works have shown that clustering these users based on the similarity of their learned preferences can significantly accelerate the learning. However, prior work has primarily focused on the online setting, which requires continually collecting user data, ignoring the offline data widely available in many applications. To tackle these limitations, we study the offline clustering of bandits (Off-ClusBand) problem, which studies how to use the offline dataset to learn cluster properties and improve decision-making across multiple users. The key challenge in Off-ClusBand arises from data insufficiency for users: unlike the online case, in the offline case, we have a fixed, limited dataset to work from and thus must determine whether we have enough data to confidently cluster users together. To address this challenge, we propose two algorithms: Off-C$^2$LUB, which we analytically show performs well for arbitrary amounts of user data, and Off-CLUB, which is prone to bias when data is limited but, given sufficient data, matches a theoretical lower bound that we derive for the offline clustered MAB problem. We experimentally validate these results on both real and synthetic datasets.

TACO: Tackling Over-correction in Federated Learning with Tailored Adaptive Correction

Non-independent and identically distributed (Non-IID) data across edge clients have long posed significant challenges to federated learning (FL) training in edge computing environments. Prior works have proposed various methods to mitigate this statistical heterogeneity. While these works can achieve good theoretical performance, in this work we provide the first investigation into a hidden over-correction phenomenon brought by the uniform model correction coefficients across clients adopted by existing methods. Such over-correction could degrade model performance and even cause failures in model convergence. To address this, we propose TACO, a novel algorithm that addresses the non-IID nature of clients' data by implementing fine-grained, client-specific gradient correction and model aggregation, steering local models towards a more accurate global optimum. Moreover, we verify that leading FL algorithms generally have better model accuracy in terms of communication rounds rather than wall-clock time, resulting from their extra computation overhead imposed on clients. To enhance the training efficiency, TACO deploys a lightweight model correction and tailored aggregation approach that requires minimum computation overhead and no extra information beyond the synchronized model parameters. To validate TACO's effectiveness, we present the first FL convergence analysis that reveals the root cause of over-correction. Extensive experiments across various datasets confirm TACO's superior and stable performance in practice.

Tin-Tin: Towards Tiny Learning on Tiny Devices with Integer-based Neural Network Training

Recent advancements in machine learning (ML) have enabled its deployment on resource-constrained edge devices, fostering innovative applications such as intelligent environmental sensing. However, these devices, particularly microcontrollers (MCUs), face substantial challenges due to limited memory, computing capabilities, and the absence of dedicated floating-point units (FPUs). These constraints hinder the deployment of complex ML models, especially those requiring lifelong learning capabilities. To address these challenges, we propose Tin-Tin, an integer-based on-device training framework designed specifically for low-power MCUs. Tin-Tin introduces novel integer rescaling techniques to efficiently manage dynamic ranges and facilitate efficient weight updates using integer data types. Unlike existing methods optimized for devices with FPUs, GPUs, or FPGAs, Tin-Tin addresses the unique demands of tiny MCUs, prioritizing energy efficiency and optimized memory utilization. We validate the effectiveness of Tin-Tin through end-to-end application examples on real-world tiny devices, demonstrating its potential to support energy-efficient and sustainable ML applications on edge platforms.

Towards Optimal Heterogeneous Client Sampling in Multi-Model Federated Learning

Federated learning (FL) allows edge devices to collaboratively train models without sharing local data. As FL gains popularity, clients may need to train multiple unrelated FL models, but communication constraints limit their ability to train all models simultaneously. While clients could train FL models sequentially, opportunistically having FL clients concurrently train different models -- termed multi-model federated learning (MMFL) -- can reduce the overall training time. Prior work uses simple client-to-model assignments that do not optimize the contribution of each client to each model over the course of its training. Prior work on single-model FL shows that intelligent client selection can greatly accelerate convergence, but na\"ive extensions to MMFL can violate heterogeneous resource constraints at both the server and the clients. In this work, we develop a novel convergence analysis of MMFL with arbitrary client sampling methods, theoretically demonstrating the strengths and limitations of previous well-established gradient-based methods. Motivated by this analysis, we propose MMFL-LVR, a loss-based sampling method that minimizes training variance while explicitly respecting communication limits at the server and reducing computational costs at the clients. We extend this to MMFL-StaleVR, which incorporates stale updates for improved efficiency and stability, and MMFL-StaleVRE, a lightweight variant suitable for low-overhead deployment. Experiments show our methods improve average accuracy by up to 19.1% over random sampling, with only a 5.4% gap from the theoretical optimum (full client participation).

Interference-Aware Edge Runtime Prediction with Conformal Matrix Completion

Accurately estimating workload runtime is a longstanding goal in computer systems, and plays a key role in efficient resource provisioning, latency minimization, and various other system management tasks. Runtime prediction is particularly important for managing increasingly complex distributed systems in which more sophisticated processing is pushed to the edge in search of better latency. Previous approaches for runtime prediction in edge systems suffer from poor data efficiency or require intensive instrumentation; these challenges are compounded in heterogeneous edge computing environments, where historical runtime data may be sparsely available and instrumentation is often challenging. Moreover, edge computing environments often feature multi-tenancy due to limited resources at the network edge, potentially leading to interference between workloads and further complicating the runtime prediction problem. Drawing from insights across machine learning and computer systems, we design a matrix factorization-inspired method that generates accurate interference-aware predictions with tight provably-guaranteed uncertainty bounds. We validate our method on a novel WebAssembly runtime dataset collected from 24 unique devices, achieving a prediction error of 5.2% -- 2x better than a naive application of existing methods.

LLM-Powered Decentralized Generative Agents with Adaptive Hierarchical Knowledge Graph for Cooperative Planning

Developing intelligent agents for long-term cooperation in dynamic open-world scenarios is a major challenge in multi-agent systems. Traditional Multi-agent Reinforcement Learning (MARL) frameworks like centralized training decentralized execution (CTDE) struggle with scalability and flexibility. They require centralized long-term planning, which is difficult without custom reward functions, and face challenges in processing multi-modal data. CTDE approaches also assume fixed cooperation strategies, making them impractical in dynamic environments where agents need to adapt and plan independently. To address decentralized multi-agent cooperation, we propose Decentralized Adaptive Knowledge Graph Memory and Structured Communication System (DAMCS) in a novel Multi-agent Crafter environment. Our generative agents, powered by Large Language Models (LLMs), are more scalable than traditional MARL agents by leveraging external knowledge and language for long-term planning and reasoning. Instead of fully sharing information from all past experiences, DAMCS introduces a multi-modal memory system organized as a hierarchical knowledge graph and a structured communication protocol to optimize agent cooperation. This allows agents to reason from past interactions and share relevant information efficiently. Experiments on novel multi-agent open-world tasks show that DAMCS outperforms both MARL and LLM baselines in task efficiency and collaboration. Compared to single-agent scenarios, the two-agent scenario achieves the same goal with 63% fewer steps, and the six-agent scenario with 74% fewer steps, highlighting the importance of adaptive memory and structured communication in achieving long-term goals. We publicly release our project at: https://happyeureka.github.io/damcs.

Pairwise Elimination with Instance-Dependent Guarantees for Bandits with Cost Subsidy

Multi-armed bandits (MAB) are commonly used in sequential online decision-making when the reward of each decision is an unknown random variable. In practice however, the typical goal of maximizing total reward may be less important than minimizing the total cost of the decisions taken, subject to a reward constraint. For example, we may seek to make decisions that have at least the reward of a reference ``default'' decision, with as low a cost as possible. This problem was recently introduced in the Multi-Armed Bandits with Cost Subsidy (MAB-CS) framework. MAB-CS is broadly applicable to problem domains where a primary metric (cost) is constrained by a secondary metric (reward), and the rewards are unknown. In our work, we address variants of MAB-CS including ones with reward constrained by the reward of a known reference arm or by the subsidized best reward. We introduce the Pairwise-Elimination (PE) algorithm for the known reference arm variant and generalize PE to PE-CS for the subsidized best reward variant. Our instance-dependent analysis of PE and PE-CS reveals that both algorithms have an order-wise logarithmic upper bound on Cost and Quality Regret, making our policies the first with such a guarantee. Moreover, by comparing our upper and lower bound results we establish that PE is order-optimal for all known reference arm problem instances. Finally, experiments are conducted using the MovieLens 25M and Goodreads datasets for both PE and PE-CS revealing the effectiveness of PE and the superior balance between performance and reliability offered by PE-CS compared to baselines from the literature.