Soft Actor-Critic (SAC) and its variants dominate Multi-Task Reinforcement Learning (MTRL) due to their off-policy sample efficiency, while on-policy methods such as Proximal Policy Optimization (PPO) remain underexplored. We diagnose that PPO in MTRL suffers from a previously overlooked issue: critic-side gradient ill-conditioning, which may cause tail tasks to stall while easy tasks dominate the value function's updates. To address this, we propose TOPPO (Tail-Optimized PPO), a reformulation of PPO via Critic Balancing -- a set of modules that improve gradient conditioning and balance learning dynamics across tasks. Unlike prior approaches that rely on modular architectures or large models, TOPPO targets the optimization bottleneck within PPO itself. Empirically, TOPPO achieves stronger mean and tail-task performance than published SAC-family and ARS-family baselines while using substantially fewer parameters and environment steps on Meta-World+ benchmark. Notably, TOPPO matches or surpasses strong SAC baselines early in training and maintains superior performance at full budget. Ablations confirm the effectiveness of each module in TOPPO and provide insights into their interactions. Our results demonstrate that, with proper optimization, on-policy methods can rival or exceed off-policy approaches in MTRL, challenging the prevailing reliance on SAC and highlighting critic-side gradient conditioning as the central bottleneck.

In sparse linear bandits, a learning agent sequentially selects an action and receive reward feedback, and the reward function depends linearly on a few coordinates of the covariates of the actions. This has applications in many real-world sequential decision making problems. In this paper, we propose a simple and computationally efficient sparse linear estimation method called POPART that enjoys a tighter ℓ1 recovery guarantee compared to Lasso (Tibshirani, 1996) in many problems. Our bound naturally motivates an experimental design criterion that is convex and thus computationally efficient to solve. Based on our novel estimator and design criterion, we derive sparse linear bandit algorithms that enjoy improved regret upper bounds upon the state of the art (Hao et al., 2020), especially w.r.t. the geometry of the given action set. Finally, we prove a matching lower bound for sparse linear bandits in the data-poor regime, which closes the gap between upper and lower bounds in prior work.

In sparse linear bandits, a learning agent sequentially selects an action from a fixed action set and receives reward feedback, and the reward function depends linearly on a few coordinates of the covariates of the actions. This has applications in many real-world sequential decision making problems. In this paper, we devise a simple, novel sparse linear estimation method called $\textrm{PopArt}$ that enjoys a tighter $\ell_1$ recovery guarantee compared to Lasso (Tibshirani, 1996). Our bound naturally motivates an experimental design criterion that is convex and thus computationally efficient to solve. Based on our novel estimator and design criterion, we derive sparse linear bandit algorithms that enjoy improved regret upper bounds upon the state of the art (Hao et al., 2020), especially in terms of the geometry of the given action set. Finally, we prove a matching lower bound for sparse linear bandits in the data-poor regime, which closes the gap between upper and lower bounds in prior work.

In sparse linear bandits, a learning agent sequentially selects an action from a fixed action set and receives reward feedback, and the reward function depends linearly on a few coordinates of the covariates of the actions. This has applications in many real-world sequential decision making problems. In this paper, we devise a simple, novel sparse linear estimation method called \textrm{PopArt} that enjoys a tighter \ell_1 recovery guarantee compared to Lasso (Tibshirani, 1996). Our bound naturally motivates an experimental design criterion that is convex and thus computationally efficient to solve. Based on our novel estimator and design criterion, we derive sparse linear bandit algorithms that enjoy improved regret upper bounds upon the state of the art (Hao et al., 2020), especially in terms of the geometry of the given action set.

Recent Offline Reinforcement Learning methods have succeeded in learning high-performance policies from fixed datasets of experience. A particularly effective approach learns to first identify and then mimic optimal decision-making strategies. Our work evaluates this method's ability to scale to vast datasets consisting almost entirely of sub-optimal noise. A thorough investigation on a custom benchmark helps identify several key challenges involved in learning from high-noise datasets. We re-purpose prioritized experience sampling to locate expert-level demonstrations among millions of low-performance samples. This modification enables offline agents to learn state-of-the-art policies in benchmark tasks using datasets where expert actions are outnumbered nearly 65:1.

The General Video Game Artificial Intelligence (GVGAI) competition has been running for several years with various tracks. This paper focuses on the challenge of the GVGAI learning track in which 3 games are selected and 2 levels are given for training, while 3 hidden levels are left for evaluation. This setup poses a difficult challenge for current Reinforcement Learning (RL) algorithms, as they typically require much more data. This work investigates 3 versions of the Advantage Actor-Critic (A2C) algorithm trained on a maximum of 2 levels from the available 5 from the GVGAI framework and compares their performance on all levels. The selected sub-set of games have different characteristics, like stochasticity, reward distribution and objectives. We found that stochasticity improves the generalisation, but too much can cause the algorithms to fail to learn the training levels. The quality of the training levels also matters, different sets of training levels can boost generalisation over all levels. In the GVGAI competition agents are scored based on their win rates and then their scores achieved in the games. We found that solely using the rewards provided by the game might not encourage winning.