$TAR^2$: Temporal-Agent Reward Redistribution for Optimal Policy Preservation in Multi-Agent Reinforcement Learning

In cooperative multi-agent reinforcement learning (MARL), learning effective policies is challenging when global rewards are sparse and delayed. This difficulty arises from the need to assign credit across both agents and time steps, a problem that existing methods often fail to address in episodic, long-horizon tasks. We propose Temporal-Agent Reward Redistribution $TAR^2$, a novel approach that decomposes sparse global rewards into agent-specific, time-step-specific components, thereby providing more frequent and accurate feedback for policy learning. Theoretically, we show that $TAR^2$ (i) aligns with potential-based reward shaping, preserving the same optimal policies as the original environment, and (ii) maintains policy gradient update directions identical to those under the original sparse reward, ensuring unbiased credit signals. Empirical results on two challenging benchmarks, SMACLite and Google Research Football, demonstrate that $TAR^2$ significantly stabilizes and accelerates convergence, outperforming strong baselines like AREL and STAS in both learning speed and final performance. These findings establish $TAR^2$ as a principled and practical solution for agent-temporal credit assignment in sparse-reward multi-agent systems.

Agent-Temporal Credit Assignment for Optimal Policy Preservation in Sparse Multi-Agent Reinforcement Learning

In multi-agent environments, agents often struggle to learn optimal policies due to sparse or delayed global rewards, particularly in long-horizon tasks where it is challenging to evaluate actions at intermediate time steps. We introduce Temporal-Agent Reward Redistribution (TAR$^2$), a novel approach designed to address the agent-temporal credit assignment problem by redistributing sparse rewards both temporally and across agents. TAR$^2$ decomposes sparse global rewards into time-step-specific rewards and calculates agent-specific contributions to these rewards. We theoretically prove that TAR$^2$ is equivalent to potential-based reward shaping, ensuring that the optimal policy remains unchanged. Empirical results demonstrate that TAR$^2$ stabilizes and accelerates the learning process. Additionally, we show that when TAR$^2$ is integrated with single-agent reinforcement learning algorithms, it performs as well as or better than traditional multi-agent reinforcement learning methods.

DeepClean: Integrated Distortion Identification and Algorithm Selection for Rectifying Image Corruptions

Distortion identification and rectification in images and videos is vital for achieving good performance in downstream vision applications. Instead of relying on fixed trial-and-error based image processing pipelines, we propose a two-level sequential planning approach for automated image distortion classification and rectification. At the higher level it detects the class of corruptions present in the input image, if any. The lower level selects a specific algorithm to be applied, from a set of externally provided candidate algorithms. The entire two-level setup runs in the form of a single forward pass during inference and it is to be queried iteratively until the retrieval of the original image. We demonstrate improvements compared to three baselines on the object detection task on COCO image dataset with rich set of distortions. The advantage of our approach is its dynamic reconfiguration, conditioned on the input image and generalisability to unseen candidate algorithms at inference time, since it relies only on the comparison of their output of the image embeddings.

Leveraging Domain Knowledge for Efficient Reward Modelling in RLHF: A Case-Study in E-Commerce Opinion Summarization

Reinforcement Learning from Human Feedback (RLHF) has become a dominating strategy in steering Language Models (LMs) towards human values/goals. The key to the strategy is employing a reward model ({$\varphi$}) which can reflect a latent reward model with humans. While this strategy has proven to be effective, the training methodology requires a lot of human preference annotation (usually of the order of tens of thousands) to train {$\varphi$}. Such large-scale preference annotations can be achievable if the reward model can be ubiquitously used. However, human values/goals are subjective and depend on the nature of the task. This poses a challenge in collecting diverse preferences for downstream applications. To address this, we propose a novel methodology to infuse domain knowledge into {$\varphi$}, which reduces the size of preference annotation required. We validate our approach in E-Commerce Opinion Summarization, with a significant reduction in dataset size (just $940$ samples) while advancing the state-of-the-art. Our contributions include a novel Reward Modelling technique, a new dataset (PromptOpinSumm) for Opinion Summarization, and a human preference dataset (OpinPref). The proposed methodology opens avenues for efficient RLHF, making it more adaptable to diverse applications with varying human values. We release the artifacts for usage under MIT License.

Transformers are Expressive, But Are They Expressive Enough for Regression?

Transformers have become pivotal in Natural Language Processing, demonstrating remarkable success in applications like Machine Translation and Summarization. Given their widespread adoption, several works have attempted to analyze the expressivity of Transformers. Expressivity of a neural network is the class of functions it can approximate. A neural network is fully expressive if it can act as a universal function approximator. We attempt to analyze the same for Transformers. Contrary to existing claims, our findings reveal that Transformers struggle to reliably approximate continuous functions, relying on piecewise constant approximations with sizable intervals. The central question emerges as: "\textit{Are Transformers truly Universal Function Approximators}?" To address this, we conduct a thorough investigation, providing theoretical insights and supporting evidence through experiments. Our contributions include a theoretical analysis pinpointing the root of Transformers' limitation in function approximation and extensive experiments to verify the limitation. By shedding light on these challenges, we advocate a refined understanding of Transformers' capabilities.

Reinforcement Replaces Supervision: Query focused Summarization using Deep Reinforcement Learning

Query-focused Summarization (QfS) deals with systems that generate summaries from document(s) based on a query. Motivated by the insight that Reinforcement Learning (RL) provides a generalization to Supervised Learning (SL) for Natural Language Generation, and thereby performs better (empirically) than SL, we use an RL-based approach for this task of QfS. Additionally, we also resolve the conflict of employing RL in Transformers with Teacher Forcing. We develop multiple Policy Gradient networks, trained on various reward signals: ROUGE, BLEU, and Semantic Similarity, which lead to a 10-point improvement over the State-of-the-Art approach on the ROUGE-L metric for a benchmark dataset (ELI5). We also show performance of our approach in zero-shot setting for another benchmark dataset (DebatePedia) -- our approach leads to results comparable to baselines, which were specifically trained on DebatePedia. To aid the RL training, we propose a better semantic similarity reward, enabled by a novel Passage Embedding scheme developed using Cluster Hypothesis. Lastly, we contribute a gold-standard test dataset to further research in QfS and Long-form Question Answering (LfQA).

Multi-Agent Learning of Efficient Fulfilment and Routing Strategies in E-Commerce

This paper presents an integrated algorithmic framework for minimising product delivery costs in e-commerce (known as the cost-to-serve or C2S). One of the major challenges in e-commerce is the large volume of spatio-temporally diverse orders from multiple customers, each of which has to be fulfilled from one of several warehouses using a fleet of vehicles. This results in two levels of decision-making: (i) selection of a fulfillment node for each order (including the option of deferral to a future time), and then (ii) routing of vehicles (each of which can carry multiple orders originating from the same warehouse). We propose an approach that combines graph neural networks and reinforcement learning to train the node selection and vehicle routing agents. We include real-world constraints such as warehouse inventory capacity, vehicle characteristics such as travel times, service times, carrying capacity, and customer constraints including time windows for delivery. The complexity of this problem arises from the fact that outcomes (rewards) are driven both by the fulfillment node mapping as well as the routing algorithms, and are spatio-temporally distributed. Our experiments show that this algorithmic pipeline outperforms pure heuristic policies.

Using General Value Functions to Learn Domain-Backed Inventory Management Policies

We consider the inventory management problem, where the goal is to balance conflicting objectives such as availability and wastage of a large range of products in a store. We propose a reinforcement learning (RL) approach that utilises General Value Functions (GVFs) to derive domain-backed inventory replenishment policies. The inventory replenishment decisions are modelled as a sequential decision making problem, which is challenging due to uncertain demand and the existence of aggregate (cross-product) constraints. In existing literature, GVFs have primarily been used for auxiliary task learning. We use this capability to train GVFs on domain-critical characteristics such as prediction of stock-out probability and wastage quantity. Using this domain expertise for more effective exploration, we train an RL agent to compute the inventory replenishment quantities for a large range of products (up to 6000 in the reported experiments), which share aggregate constraints such as the total weight/volume per delivery. Additionally, we show that the GVF predictions can be used to provide additional domain-backed insights into the decisions proposed by the RL agent. Finally, since the environment dynamics are fully transferred, the trained GVFs can be used for faster adaptation to vastly different business objectives (for example, due to the start of a promotional period or due to deployment in a new customer environment).

Using Linear Regression for Iteratively Training Neural Networks

We present a simple linear regression based approach for learning the weights and biases of a neural network, as an alternative to standard gradient based backpropagation. The present work is exploratory in nature, and we restrict the description and experiments to (i) simple feedforward neural networks, (ii) scalar (single output) regression problems, and (iii) invertible activation functions. However, the approach is intended to be extensible to larger, more complex architectures. The key idea is the observation that the input to every neuron in a neural network is a linear combination of the activations of neurons in the previous layer, as well as the parameters (weights and biases) of the layer. If we are able to compute the ideal total input values to every neuron by working backwards from the output, we can formulate the learning problem as a linear least squares problem which iterates between updating the parameters and the activation values. We present an explicit algorithm that implements this idea, and we show that (at least for small problems) the approach is more stable and faster than gradient-based methods.

DCT: Dual Channel Training of Action Embeddings for Reinforcement Learning with Large Discrete Action Spaces

The ability to learn robust policies while generalizing over large discrete action spaces is an open challenge for intelligent systems, especially in noisy environments that face the curse of dimensionality. In this paper, we present a novel framework to efficiently learn action embeddings that simultaneously allow us to reconstruct the original action as well as to predict the expected future state. We describe an encoder-decoder architecture for action embeddings with a dual channel loss that balances between action reconstruction and state prediction accuracy. We use the trained decoder in conjunction with a standard reinforcement learning algorithm that produces actions in the embedding space. Our architecture is able to outperform two competitive baselines in two diverse environments: a 2D maze environment with more than 4000 discrete noisy actions, and a product recommendation task that uses real-world e-commerce transaction data. Empirical results show that the model results in cleaner action embeddings, and the improved representations help learn better policies with earlier convergence.