STEERING: Stein Information Directed Exploration for Model-Based Reinforcement Learning

Directed Exploration is a crucial challenge in reinforcement learning (RL), especially when rewards are sparse. Information-directed sampling (IDS), which optimizes the information ratio, seeks to do so by augmenting regret with information gain. However, estimating information gain is computationally intractable or relies on restrictive assumptions which prohibit its use in many practical instances. In this work, we posit an alternative exploration incentive in terms of the integral probability metric (IPM) between a current estimate of the transition model and the unknown optimal, which under suitable conditions, can be computed in closed form with the kernelized Stein discrepancy (KSD). Based on KSD, we develop a novel algorithm STEERING: \textbf{STE}in information dir\textbf{E}cted exploration for model-based \textbf{R}einforcement Learn\textbf{ING}. To enable its derivation, we develop fundamentally new variants of KSD for discrete conditional distributions. We further establish that STEERING archives sublinear Bayesian regret, improving upon prior learning rates of information-augmented MBRL, IDS included. Experimentally, we show that the proposed algorithm is computationally affordable and outperforms several prior approaches.

Synthetic Wave-Geometric Impulse Responses for Improved Speech Dereverberation

We present a novel approach to improve the performance of learning-based speech dereverberation using accurate synthetic datasets. Our approach is designed to recover the reverb-free signal from a reverberant speech signal. We show that accurately simulating the low-frequency components of Room Impulse Responses (RIRs) is important to achieving good dereverberation. We use the GWA dataset that consists of synthetic RIRs generated in a hybrid fashion: an accurate wave-based solver is used to simulate the lower frequencies and geometric ray tracing methods simulate the higher frequencies. We demonstrate that speech dereverberation models trained on hybrid synthetic RIRs outperform models trained on RIRs generated by prior geometric ray tracing methods on four real-world RIR datasets.

Towards Improved Room Impulse Response Estimation for Speech Recognition

We propose to characterize and improve the performance of blind room impulse response (RIR) estimation systems in the context of a downstream application scenario, far-field automatic speech recognition (ASR). We first draw the connection between improved RIR estimation and improved ASR performance, as a means of evaluating neural RIR estimators. We then propose a GAN-based architecture that encodes RIR features from reverberant speech and constructs an RIR from the encoded features, and uses a novel energy decay relief loss to optimize for capturing energy-based properties of the input reverberant speech. We show that our model outperforms the state-of-the-art baselines on acoustic benchmarks (by 72% on the energy decay relief and 22% on an early-reflection energy metric), as well as in an ASR evaluation task (by 6.9% in word error rate).

SLICER: Learning universal audio representations using low-resource self-supervised pre-training

We present a new Self-Supervised Learning (SSL) approach to pre-train encoders on unlabeled audio data that reduces the need for large amounts of labeled data for audio and speech classification. Our primary aim is to learn audio representations that can generalize across a large variety of speech and non-speech tasks in a low-resource un-labeled audio pre-training setting. Inspired by the recent success of clustering and contrasting learning paradigms for SSL-based speech representation learning, we propose SLICER (Symmetrical Learning of Instance and Cluster-level Efficient Representations), which brings together the best of both clustering and contrasting learning paradigms. We use a symmetric loss between latent representations from student and teacher encoders and simultaneously solve instance and cluster-level contrastive learning tasks. We obtain cluster representations online by just projecting the input spectrogram into an output subspace with dimensions equal to the number of clusters. In addition, we propose a novel mel-spectrogram augmentation procedure, k-mix, based on mixup, which does not require labels and aids unsupervised representation learning for audio. Overall, SLICER achieves state-of-the-art results on the LAPE Benchmark \cite{9868132}, significantly outperforming DeLoRes-M and other prior approaches, which are pre-trained on $10\times$ larger of unsupervised data. We will make all our codes available on GitHub.

MAST: Multiscale Audio Spectrogram Transformers

We present Multiscale Audio Spectrogram Transformer (MAST) for audio classification, which brings the concept of multiscale feature hierarchies to the Audio Spectrogram Transformer (AST). Given an input audio spectrogram we first patchify and project it into an initial temporal resolution and embedding dimension, post which the multiple stages in MAST progressively expand the embedding dimension while reducing the temporal resolution of the input. We use a pyramid structure that allows early layers of MAST operating at a high temporal resolution but low embedding space to model simple low-level acoustic information and deeper temporally coarse layers to model high-level acoustic information with high-dimensional embeddings. We also extend our approach to present a new Self-Supervised Learning (SSL) method called SS-MAST, which calculates a symmetric contrastive loss between latent representations from a student and a teacher encoder. In practice, MAST significantly outperforms AST by an average accuracy of 3.4% across 8 speech and non-speech tasks from the LAPE Benchmark. Moreover, SS-MAST achieves an absolute average improvement of 2.6% over SSAST for both AST and MAST encoders. We make all our codes available on GitHub at the time of publication.

WGICP: Differentiable Weighted GICP-Based Lidar Odometry

We present a novel differentiable weighted generalized iterative closest point (WGICP) method applicable to general 3D point cloud data, including that from Lidar. Our method builds on differentiable generalized ICP (GICP), and we propose using the differentiable K-Nearest Neighbor (KNN) algorithm to enhance differentiability. The differentiable GICP algorithm provides the gradient of output pose estimation with respect to each input point, which allows us to train a neural network to predict its importance, or weight, in estimating the correct pose. In contrast to the other ICP-based methods that use voxel-based downsampling or matching methods to reduce the computational cost, our method directly reduces the number of points used for GICP by only selecting those with the highest weights and ignoring redundant ones with lower weights. We show that our method improves both accuracy and speed of the GICP algorithm for the KITTI dataset and can be used to develop a more robust and efficient SLAM system.

DMCA: Dense Multi-agent Navigation using Attention and Communication

In decentralized multi-robot navigation, the agents lack the world knowledge to make safe and (near-)optimal plans reliably and make their decisions on their neighbors' observable states. We present a reinforcement learning based multi-agent navigation algorithm that performs inter-agent communications. In order to deal with the variable number of neighbors for each agent, we use a multi-head self-attention mechanism to encode neighbor information and create a fixed-length observation vector. We pose communication selection as a link prediction problem, where the network predicts whether communication is necessary given the observable information. The communicated information augments the observed neighbor information and is used to select a suitable navigation plan. We highlight the benefits of our approach by performing safe and efficient navigation among multiple robots in dense and challenging benchmarks. We also compare the performance with other learning-based methods and highlight improvements in terms of fewer collisions and time-to-goal in dense scenarios.

MSVIPER: Improved Policy Distillation for Reinforcement-Learning-Based Robot Navigation

We present Multiple Scenario Verifiable Reinforcement Learning via Policy Extraction (MSVIPER), a new method for policy distillation to decision trees for improved robot navigation. MSVIPER learns an "expert" policy using any Reinforcement Learning (RL) technique involving learning a state-action mapping and then uses imitation learning to learn a decision-tree policy from it. We demonstrate that MSVIPER results in efficient decision trees and can accurately mimic the behavior of the expert policy. Moreover, we present efficient policy distillation and tree-modification techniques that take advantage of the decision tree structure to allow improvements to a policy without retraining. We use our approach to improve the performance of RL-based robot navigation algorithms for indoor and outdoor scenes. We demonstrate the benefits in terms of reduced freezing and oscillation behaviors (by up to 95\% reduction) for mobile robots navigating among dynamic obstacles and reduced vibrations and oscillation (by up to 17\%) for outdoor robot navigation on complex, uneven terrains.

Differentiable Frequency-based Disentanglement for Aerial Video Action Recognition

We present a learning algorithm for human activity recognition in videos. Our approach is designed for UAV videos, which are mainly acquired from obliquely placed dynamic cameras that contain a human actor along with background motion. Typically, the human actors occupy less than one-tenth of the spatial resolution. Our approach simultaneously harnesses the benefits of frequency domain representations, a classical analysis tool in signal processing, and data driven neural networks. We build a differentiable static-dynamic frequency mask prior to model the salient static and dynamic pixels in the video, crucial for the underlying task of action recognition. We use this differentiable mask prior to enable the neural network to intrinsically learn disentangled feature representations via an identity loss function. Our formulation empowers the network to inherently compute disentangled salient features within its layers. Further, we propose a cost-function encapsulating temporal relevance and spatial content to sample the most important frame within uniformly spaced video segments. We conduct extensive experiments on the UAV Human dataset and the NEC Drone dataset and demonstrate relative improvements of 5.72% - 13.00% over the state-of-the-art and 14.28% - 38.05% over the corresponding baseline model.