Searcharxiv⌕ Search

arXiv subjects

Jiin Woo

Publications and source records attributed to Jiin Woo.

12 recordsLinked to original sources

Exploration via Exploitation: The Blessing of Reward Diversity in Personalized Federated RL

While federated reinforcement learning (RL) has traditionally guaranteed high efficiency in homogeneous environments, collaboration is often hindered when agents possess distinct goals. In this paper, we flip this paradigm, framing reward heterogeneity not as a hurdle to consensus, but as a structural asset that catalyzes exploration. We first propose Personalized Federated Upper Confidence Bound Value Iteration (PF-UCBVI), which achieves optimal linear speedup by decoupling shared dynamics from personalized objectives. To eliminate the risks of explicit exploration, we then introduce Personalized Federated Exploration-Free Value Iteration (PF-EFVI), a purely greedy algorithm that leverages reward diversity to ensure state-action coverage. We prove that PF-EFVI attains logarithmic regret without explicit exploration bonuses under sufficient reward diversity. Our results reveal that agent disagreement shifts the exploration burden from temporal complexity to spatial diversity, enabling efficient collective learning without explicit exploration.

cs.LG↗

Compatibility Matters: A Framework for Token-Level Teacher Selection in On-Policy Distillation

On-policy distillation (OPD) has recently been introduced as a promising paradigm for continual learning, where a student policy can be supervised at the token-level. Typically, OPD is performed using a single teacher per prompt rollout (or even per training run). However, this ignores a whole other axis of flexibility that is available due to the on-policy nature - token-level teacher selection during rollouts. Building on this insight, we theoretically explore what makes a teacher and a student compatible for OPD, and leverage our findings to derive several empirical heuristics for teacher selection. In particular, our framework captures several desiderata for OPD: compatible teacher-student reasoning patterns, teacher competence, and induced student diversity. Theoretically, we categorize several types of teacher routing schemes, and we prove that under different scenarios, different routing schemes are likely to be beneficial for optimization and student diversity. We back up our findings with extensive experimental results in domains such as math, coding, kernel generation, and visual reasoning, where we find upwards of 10 percentage points of improvement on pass@$k$ when compared to vanilla OPD and the teacher experts on certain tasks, indicating both capability improvement and diversity over high-quality responses.

cs.LG↗

Multi-Reward Hacking: From Aggregation-Induced Drift to Trust-Anchored Defense

To improve task performance while encouraging desirable response properties, LLM post-training often combines a primary reward (e.g., correctness) with auxiliary rewards (e.g., brevity or efficiency). However, how these rewards are aggregated can critically affect training outcomes. We show that symmetric aggregation, which treats reward channels equally, can induce a failure where aggregate reward improves while primary-task performance degrades, even when individual rewards are meaningful. We investigate this aggregation-induced failure through a one-step multi-reward decision formulation, characterizing when it constitutes multi-reward hacking. We prove that, under symmetric aggregation, deterministic conditional-mean GRPO converges to auxiliary-favoring behavior when auxiliary rewards are more reliably obtained, regardless of initialization. We propose Trust-Anchored Gating (TAG), which first elects an anchor channel based on the policy drift induced by optimizing each channel alone, then gates auxiliary credit on anchor satisfaction. Unlike calibration-sensitive static reweighting, TAG structurally eliminates credit for auxiliary-only outcomes while preserving additional credit for joint success. In the same analytical setting, TAG provably reverses this selection, converging to the primary-favoring action from any initialization when the primary channel is anchored. Experiments across Qwen and Llama models on mathematics and program synthesis reveal substantial degradation under symmetric, constrained, and normalized aggregation, whereas TAG mitigates this failure. On Qwen3-4B with a brevity reward, TAG achieves 0.581 pass@4 versus 0.220 for additive aggregation, while shortening responses by 44% relative to correctness-only training.

cs.LG↗

How to Train Your Little Dragon: Teaching a Small Language Model to Construct and Refine GPU Optimization Scripts

Compiler transformations interact, making GPU kernel optimization a search for combinations that preserve the computation while improving performance. We introduce HOW TO TRAIN YOUR LITTLE DRAGON, a training method that teaches a small language model both to construct complete optimization scripts and to improve existing ones. Instead of ranking candidates or generating kernel code, our model generates optimization scripts. We devise a compact, lossless representation whose deterministic decoding reconstructs the exact compiler trace, preserving its transformation parameters and dependencies. We use a compiler framework TVM MetaSchedule to build training data on 54,116 distinct workloads, and construct 108,594 examples in two training sets: 60,236 teach the model to generate an optimized script from an unoptimized kernel, while 48,358 teach it to produce a better script given the kernel and a somewhat optimized script. Supervised fine-tuning teaches the model to construct and improve optimization scripts from these search-derived examples. We then add reinforcement learning with a shaped reward that combines correctness with measured performance, awarding a latency-based bonus only to numerically correct kernels. We apply this training method to a model with 8 billion parameters. After supervised fine-tuning and reinforcement learning, the model achieves per-attempt correctness of 91.27% for construction and 92.01% for refinement, while reinforcement learning produces faster kernels than supervised training on 88 of 102 construction tasks and 145 of 170 refinement prompts. With only 8B parameters, our LITTLE DRAGON achieves higher per-attempt correctness and better kernel performance than six frontier LLMs: GPT-6 Astra, Opus 5, Kimi-K3, and GLM-5.3 among them, each with hundreds of billions to trillions of parameters, for which task-specific training is unavailable or expensive.

cs.LG↗

Federate the Router: Learning Language Model Routers with Sparse and Decentralized Evaluations

Large language models (LLMs) are increasingly accessed as remotely hosted services by edge and enterprise clients that cannot run frontier models locally. Since models vary widely in capability and price, routing queries to models that balance quality and inference cost is essential. Existing router approaches assume access to centralized query-model evaluation data. However, these data are often fragmented across clients, such as end users and organizations, and are privacy-sensitive, which makes centralizing data infeasible. Additionally, per-client router training is ineffective since local evaluation data is limited and covers only a restricted query distribution and a biased subset of model evaluations. We introduce the first federated framework for LLM routing, enabling clients to learn a shared routing policy from local offline query-model evaluation data. Our framework supports both parametric multilayer perceptron router and nonparametric K-means router under heterogeneous client query distributions and non-uniform model coverage. Across three benchmarks, federated collaboration improves the accuracy-cost frontier over client-local routers, both via increased effective model coverage and better query generalization. Our theoretical results also validate that federated training reduces routing suboptimality.

cs.LG↗

TritonRL: Training LLMs to Think and Code Triton Without Cheating

The rapid evolution of Large Language Models (LLMs) has driven a growing demand for automated, high-performance system kernels to accelerate machine learning workloads. We introduce TritonRL, a domain-specialized 8B-scale LLM for Triton programming, trained via a novel reinforcement learning (RL) framework. While Triton synthesis faces unique challenges, including data scarcity and a high susceptibility to reward hacking, our approach enables robust kernel generation through two primary innovations. First, we implement a multi-layered verification system that provides high-fidelity reward signals, ensuring that generated kernels are both syntactically and functionally valid. Second, we propose Hierarchical Reward Decomposition (HRD), which decouples reinforcement for high-level reasoning and low-level implementation to resolve the credit assignment problem in long-sequence generation. Comprehensive evaluations on KernelBench demonstrate that TritonRL achieves state-of-the-art correctness and runtime speedup, outperforming concurrent Triton-specific models and matching the performance of frontier models with over 100B parameters. Our results highlight the effectiveness of hardware-aware RL paradigms in specialized domain adaptation.

cs.SE↗

Sample Complexity of Average-Reward Q-Learning: From Single-agent to Federated Reinforcement Learning

Average-reward reinforcement learning offers a principled framework for long-term decision-making by maximizing the mean reward per time step. Although Q-learning is a widely used model-free algorithm with established sample complexity in discounted and finite-horizon Markov decision processes (MDPs), its theoretical guarantees for average-reward settings remain limited. This work studies a simple but effective Q-learning algorithm for average-reward MDPs with finite state and action spaces under the weakly communicating assumption, covering both single-agent and federated scenarios. For the single-agent case, we show that Q-learning with carefully chosen parameters achieves sample complexity $\widetilde{O}\left(\frac{|\mathcal{S}||\mathcal{A}|\|h^{\star}\|_{\mathsf{sp}}^3}{\varepsilon^3}\right)$, where $\|h^{\star}\|_{\mathsf{sp}}$ is the span norm of the bias function, improving previous results by at least a factor of $\frac{\|h^{\star}\|_{\mathsf{sp}}^2}{\varepsilon^2}$. In the federated setting with $M$ agents, we prove that collaboration reduces the per-agent sample complexity to $\widetilde{O}\left(\frac{|\mathcal{S}||\mathcal{A}|\|h^{\star}\|_{\mathsf{sp}}^3}{M\varepsilon^3}\right)$, with only $\widetilde{O}\left(\frac{\|h^{\star}\|_{\mathsf{sp}}}{\varepsilon}\right)$ communication rounds required. These results establish the first federated Q-learning algorithm for average-reward MDPs, with provable efficiency in both sample and communication complexity.

stat.ML↗

Large Language Model-Enhanced Reinforcement Learning for Diverse and Novel Recommendations

In recommendation systems, diversity and novelty are essential for capturing varied user preferences and encouraging exploration, yet many systems prioritize click relevance. While reinforcement learning (RL) has been explored to improve diversity, it often depends on random exploration that may not align with user interests. We propose LAAC (LLM-guided Adversarial Actor Critic), a novel method that leverages large language models (LLMs) as reference policies to suggest novel items, while training a lightweight policy to refine these suggestions using system-specific data. The method formulates training as a bilevel optimization between actor and critic networks, enabling the critic to selectively favor promising novel actions and the actor to improve its policy beyond LLM recommendations. To mitigate overestimation of unreliable LLM suggestions, we apply regularization that anchors critic values for unexplored items close to well-estimated dataset actions. Experiments on real-world datasets show that LAAC outperforms existing baselines in diversity, novelty, and accuracy, while remaining robust on imbalanced data, effectively integrating LLM knowledge without expensive fine-tuning.

cs.LG↗

Federated Offline Reinforcement Learning: Collaborative Single-Policy Coverage Suffices

Offline reinforcement learning (RL), which seeks to learn an optimal policy using offline data, has garnered significant interest due to its potential in critical applications where online data collection is infeasible or expensive. This work explores the benefit of federated learning for offline RL, aiming at collaboratively leveraging offline datasets at multiple agents. Focusing on finite-horizon episodic tabular Markov decision processes (MDPs), we design FedLCB-Q, a variant of the popular model-free Q-learning algorithm tailored for federated offline RL. FedLCB-Q updates local Q-functions at agents with novel learning rate schedules and aggregates them at a central server using importance averaging and a carefully designed pessimistic penalty term. Our sample complexity analysis reveals that, with appropriately chosen parameters and synchronization schedules, FedLCB-Q achieves linear speedup in terms of the number of agents without requiring high-quality datasets at individual agents, as long as the local datasets collectively cover the state-action space visited by the optimal policy, highlighting the power of collaboration in the federated setting. In fact, the sample complexity almost matches that of the single-agent counterpart, as if all the data are stored at a central location, up to polynomial factors of the horizon length. Furthermore, FedLCB-Q is communication-efficient, where the number of communication rounds is only linear with respect to the horizon length up to logarithmic factors.

cs.LG↗

The Blessing of Heterogeneity in Federated Q-Learning: Linear Speedup and Beyond

When the data used for reinforcement learning (RL) are collected by multiple agents in a distributed manner, federated versions of RL algorithms allow collaborative learning without the need for agents to share their local data. In this paper, we consider federated Q-learning, which aims to learn an optimal Q-function by periodically aggregating local Q-estimates trained on local data alone. Focusing on infinite-horizon tabular Markov decision processes, we provide sample complexity guarantees for both the synchronous and asynchronous variants of federated Q-learning. In both cases, our bounds exhibit a linear speedup with respect to the number of agents and near-optimal dependencies on other salient problem parameters. In the asynchronous setting, existing analyses of federated Q-learning, which adopt an equally weighted averaging of local Q-estimates, require that every agent covers the entire state-action space. In contrast, our improved sample complexity scales inverse proportionally to the minimum entry of the average stationary state-action occupancy distribution of all agents, thus only requiring the agents to collectively cover the entire state-action space, unveiling the blessing of heterogeneity in enabling collaborative learning by relaxing the coverage requirement of the single-agent case. However, its sample complexity still suffers when the local trajectories are highly heterogeneous. In response, we propose a novel federated Q-learning algorithm with importance averaging, giving larger weights to more frequently visited state-action pairs, which achieves a robust linear speedup as if all trajectories are centrally processed, regardless of the heterogeneity of local behavior policies.

cs.LG↗

Rumor Source Detection under Querying with Untruthful Answers

Social networks are the major routes for most individuals to exchange their opinions about new products, social trends and political issues via their interactions. It is often of significant importance to figure out who initially diffuses the information, ie, finding a rumor source or a trend setter. It is known that such a task is highly challenging and the source detection probability cannot be beyond 31 percent for regular trees, if we just estimate the source from a given diffusion snapshot. In practice, finding the source often entails the process of querying that asks "Are you the rumor source?" or "Who tells you the rumor?" that would increase the chance of detecting the source. In this paper, we consider two kinds of querying: (a) simple batch querying and (b) interactive querying with direction under the assumption that queries can be untruthful with some probability. We propose estimation algorithms for those queries, and quantify their detection performance and the amount of extra budget due to untruthfulness, analytically showing that querying significantly improves the detection performance. We perform extensive simulations to validate our theoretical findings over synthetic and real-world social network topologies.

cs.SI↗

Information Source Finding in Networks: Querying with Budgets

In this paper, we study a problem of detecting the source of diffused information by querying individuals, given a sample snapshot of the information diffusion graph, where two queries are asked: {\em (i)} whether the respondent is the source or not, and {\em (ii)} if not, which neighbor spreads the information to the respondent. We consider the case when respondents may not always be truthful and some cost is taken for each query. Our goal is to quantify the necessary and sufficient budgets to achieve the detection probability $1-δ$ for any given $0<δ<1.$ To this end, we study two types of algorithms: adaptive and non-adaptive ones, each of which corresponds to whether we adaptively select the next respondents based on the answers of the previous respondents or not. We first provide the information theoretic lower bounds for the necessary budgets in both algorithm types. In terms of the sufficient budgets, we propose two practical estimation algorithms, each of non-adaptive and adaptive types, and for each algorithm, we quantitatively analyze the budget which ensures $1-δ$ detection accuracy. This theoretical analysis not only quantifies the budgets needed by practical estimation algorithms achieving a given target detection accuracy in finding the diffusion source, but also enables us to quantitatively characterize the amount of extra budget required in non-adaptive type of estimation, refereed to as {\em adaptivity gap}. We validate our theoretical findings over synthetic and real-world social network topologies.

cs.SI↗