Publications

Mixtures of Experts Unlock Parameter Scaling for Deep RL

Johan Samir Obando Ceron

Ghada Sokar

Timon Willi

Clare Lyle

Jesse Farebrother

Jakob Nicolaus Foerster

Doina Precup

Pablo Samuel Castro

2024-02-13

ArXiv (preprint)

Mixtures of Experts Unlock Parameter Scaling for Deep RL

Johan Samir Obando Ceron

Ghada Sokar

Timon Willi

Clare Lyle

Jesse Farebrother

Jakob Nicolaus Foerster

Doina Precup

Pablo Samuel Castro

The recent rapid progress in (self) supervised learning models is in large part predicted by empirical scaling laws: a model's performance s… (see more)cales proportionally to its size. Analogous scaling laws remain elusive for reinforcement learning domains, however, where increasing the parameter count of a model often hurts its final performance. In this paper, we demonstrate that incorporating Mixture-of-Expert (MoE) modules, and in particular Soft MoEs (Puigcerver et al., 2023), into value-based networks results in more parameter-scalable models, evidenced by substantial performance increases across a variety of training regimes and model sizes. This work thus provides strong empirical evidence towards developing scaling laws for reinforcement learning.

2024-02-13

ArXiv (preprint)

Mixtures of Experts Unlock Parameter Scaling for Deep RL

Johan Samir Obando Ceron

Ghada Sokar

Timon Willi

Clare Lyle

Jesse Farebrother

Jakob Nicolaus Foerster

Doina Precup

Pablo Samuel Castro

The recent rapid progress in (self) supervised learning models is in large part predicted by empirical scaling laws: a model's performance s… (see more)cales proportionally to its size. Analogous scaling laws remain elusive for reinforcement learning domains, however, where increasing the parameter count of a model often hurts its final performance. In this paper, we demonstrate that incorporating Mixture-of-Expert (MoE) modules, and in particular Soft MoEs (Puigcerver et al., 2023), into value-based networks results in more parameter-scalable models, evidenced by substantial performance increases across a variety of training regimes and model sizes. This work thus provides strong empirical evidence towards developing scaling laws for reinforcement learning.

2024-02-13

ArXiv (preprint)

Model approximation in MDPs with unbounded per-step cost

Berk Bozkurt

Aditya Mahajan

Ashutosh Nayyar

Yi Ouyang

We consider the problem of designing a control policy for an infinite-horizon discounted cost Markov decision process …

2024-02-13

ArXiv (preprint)

Regional Adaptive Metropolis Light Transport

Hisanari Otsu

Killian Herveau

Johannes Hanika

Derek Nowrouzezahrai

Carsten Dachsbacher

The design of the proposal distributions, and most notably the kernel parameters, are crucial for the performance of Markov chain Monte Carl… (see more)o (MCMC) rendering. A poor selection of parameters can increase the correlation of the Markov chain and result in bad rendering performance. We approach this problem by a novel path perturbation strategy for online-learning of state-dependent kernel parameters. We base our approach on the theoretical framework of regional adaptive MCMC which enables the adaptation of parameters depending on the region of the state space which contains the current sample, and on information collected from previous samples. For this, we define a partitioning of the path space on a low-dimensional canonical space to capture the characteristics of paths, with a focus on path segments closer to the sensor. Fast convergence is achieved by adaptive refinement of the partitions. Exemplarily, we present two novel regional adaptive path perturbation techniques akin to lens and multi-chain perturbations. Our approach can easily be used on top of existing path space MLT methods to improve rendering efficiency, while being agnostic to the initial choice of kernel parameters.

2024-02-13

ArXiv (preprint)

Antagonistic AI

Alice Cai

Elena L. Glassman

The vast majority of discourse around AI development assumes that subservient,"moral"models aligned with"human values"are universally benefi… (see more)cial -- in short, that good AI is sycophantic AI. We explore the shadow of the sycophantic paradigm, a design space we term antagonistic AI: AI systems that are disagreeable, rude, interrupting, confrontational, challenging, etc. -- embedding opposite behaviors or values. Far from being"bad"or"immoral,"we consider whether antagonistic AI systems may sometimes have benefits to users, such as forcing users to confront their assumptions, build resilience, or develop healthier relational boundaries. Drawing from formative explorations and a speculative design workshop where participants designed fictional AI technologies that employ antagonism, we lay out a design space for antagonistic AI, articulating potential benefits, design techniques, and methods of embedding antagonistic elements into user experience. Finally, we discuss the many ethical challenges of this space and identify three dimensions for the responsible design of antagonistic AI -- consent, context, and framing.

2024-02-12

ArXiv (preprint)

Antagonistic AI

Alice Cai

Elena L. Glassman

2024-02-12

ArXiv (preprint)

Antagonistic AI

Alice Cai

Elena L. Glassman

2024-02-12

ArXiv (preprint)

Antagonistic AI

Alice Cai

Elena L. Glassman

The vast majority of discourse around AI development assumes that subservient,"moral"models aligned with"human values"are universally benefi… (see more)cial -- in short, that good AI is sycophantic AI. We explore the shadow of the sycophantic paradigm, a design space we term antagonistic AI: AI systems that are disagreeable, rude, interrupting, confrontational, challenging, etc. -- embedding opposite behaviors or values. Far from being"bad"or"immoral,"we consider whether antagonistic AI systems may sometimes have benefits to users, such as forcing users to confront their assumptions, build resilience, or develop healthier relational boundaries. Drawing from formative explorations and a speculative design workshop where participants designed fictional AI technologies that employ antagonism, we lay out a design space for antagonistic AI, articulating potential benefits, design techniques, and methods of embedding antagonistic elements into user experience. Finally, we discuss the many ethical challenges of this space and identify three dimensions for the responsible design of antagonistic AI -- consent, context, and framing.

2024-02-12

ArXiv (preprint)

An enhanced wideband tracking method for characteristic modes

Chao Huang

Chenjiang Guo

Xia Ma

Yi Yuan

Jun Ding

An enhanced wideband tracking method for characteristic modes (CMs) is investigated in this paper. The method consists of three stages, and … (see more)its core tracking stage (CTS) is based on a classical eigenvector correlation-based algorithm. To decrease the tracking time and eliminate the crossing avoidance (CRA), we append a commonly used eigenvalue filter (EF) as the preprocessing stage and a novel postprocessing stage to the CTS. The proposed postprocessing stage can identify all CRA mode pairs by analyzing their trajectory and correlation characteristics. Subsequently, it can predict corresponding CRA frequencies and correct problematic qualities rapidly. Considering potential variations in eigenvector numbers at consecutive frequency samples caused by the EF, a new execution condition for the adaptive frequency adjustment in the CTS is introduced. Finally, CMs of a conductor plate and a fractal structure are investigated to demonstrate the performance of the proposed method, and the obtained results are discussed.

2024-02-12

International Journal of Microwave and Wireless Technologies (published)

Leveraging Function Space Aggregation for Federated Learning at Scale

Nikita Dhawan

Nicole Elyse Mitchell

Zachary Charles

Zachary Garrett

The federated learning paradigm has motivated the development of methods for aggregating multiple client updates into a global server model,… (see more) without sharing client data. Many federated learning algorithms, including the canonical Federated Averaging (FedAvg), take a direct (possibly weighted) average of the client parameter updates, motivated by results in distributed optimization. In this work, we adopt a function space perspective and propose a new algorithm, FedFish, that aggregates local approximations to the functions learned by clients, using an estimate based on their Fisher information. We evaluate FedFish on realistic, large-scale cross-device benchmarks. While the performance of FedAvg can suffer as client models drift further apart, we demonstrate that FedFish is more robust to longer local training. Our evaluation across several settings in image and language benchmarks shows that FedFish outperforms FedAvg as local training epochs increase. Further, FedFish results in global networks that are more amenable to efficient personalization via local fine-tuning on the same or shifted data distributions. For instance, federated pretraining on the C4 dataset, followed by few-shot personalization on Stack Overflow, results in a 7% improvement in next-token prediction by FedFish over FedAvg.

2024-02-12

TMLR (accepted)