Search

Contemporary radio access networks employ link adaption (LA) algorithms to optimize the modulation and coding schemes to adapt to the prevailing propagation conditions and are near-optimal in terms of the achieved spectral efficiency. LA is a challenging task in the presence of mobility, fast fading, and imperfect channel quality information and limited knowledge of the receiver characteristics at the transmitter, which render model-based LA algorithms complex and suboptimal. Model-based LA is especially difficult as connected user equipment devices become increasingly heterogeneous in terms of receiver capabilities, antenna configurations and hardware characteristics. Recognizing these difficulties, previous works have proposed reinforcement learning (RL) for LA, which faces deployment difficulties due to their potential negative impacts on live performance. To address this challenge, this paper considers offline RL to learn LA policies from data acquired in live networks with minimal or no intrusive effects on the network operation. We propose three LA designs based on batch-constrained deep Q-learning, conservative Q-learning, and decision transformers, showing that offline RL algorithms can achieve performance of state-of-the-art online RL methods when data is collected with a proper behavioral policy.

Blockchain technology is booming up the digital world in recent days and thus paved a way for creating separate blockchain network for various industries. This technology is characterized by its distributed, decentralized, and immutable ledger system which serves as a fundamental platform for managing smart contract transactions (SCTs). However, these self-executing codes implemented using blockchains undergo sequential validation within a block which introduces performance bottlenecks. In response, this paper introduces a framework called the Multi-Bin Parallel Scheduler (MBPS) designed for parallelizing blockchain smart contract transactions to leverage the capabilities of multicore systems. Our proposed framework facilitates concurrent execution of SCTs, enhancing performance by allowing non-conflicting transactions to be processed simultaneously while preserving deterministic order. The framework comprises of three vital stages: conflict detection, bin creation and execution. We conducted an evaluation of our MBPS framework in Hyperledger Sawtooth v1.2.6, revealing substantial performance enhancements compared to existing parallel SCT execution frameworks across various smart contract applications. This research contributes to the ongoing optimization efforts in blockchain technology demonstrating its potential for scalability and efficiency in real-world scenarios.

Recent advances in generative modeling with diffusion processes (DPs) enabled breakthroughs in image synthesis. Despite impressive image quality, these models have various prompt compliance problems, including low recall in generating multiple objects, difficulty in generating text in images, and meeting constraints like object locations and pose. For fine-grained editing and manipulation, they also require fine-grained semantic or instance maps that are tedious to produce manually. While prompt compliance can be enhanced by addition of loss functions at inference, this is time consuming and does not scale to complex scenes. To overcome these limitations, this work introduces a new family of \textit{Factor Graph Diffusion Models} (FG-DMs) that models the joint distribution of images and conditioning variables, such as semantic, sketch, depth or normal maps via a factor graph decomposition. This joint structure has several advantages, including support for efficient sampling based prompt compliance schemes, which produce images of high object recall, semi-automated fine-grained editing, text-based editing of conditions with noise inversion, explainability at intermediate levels, ability to produce labeled datasets for the training of downstream models such as segmentation or depth, training with missing data, and continual learning where new conditioning variables can be added with minimal or no modifications to the existing structure. We propose an implementation of FG-DMs by adapting a pre-trained Stable Diffusion (SD) model to implement all FG-DM factors, using only COCO dataset, and show that it is effective in generating images with 15\% higher recall than SD while retaining its generalization ability. We introduce an attention distillation loss that encourages consistency among the attention maps of all factors, improving the fidelity of the generated conditions and image., Comment: Accepted to NeurIPS 2024 conference. Project Page: https://deepaksridhar.github.io/factorgraphdiffusion.github.io/

The ability for optically active media to rotate the polarization of light is the basis of polarimetry, an illustrious technique responsible for many breakthroughs in fields as varied as astronomy, medicine and material science. Here, we recast the primary mechanism for spin readout in semiconductor-based quantum computers, Pauli spin-blockade (PSB), as the natural extension of polarimetry to the third dimension. We perform polarimetry with spins through a silicon quantum dot exchanging a hole with a boron acceptor, demonstrating the role of spin-orbit coupling in creating spin misalignment. Perfect spin alignment may be recovered by means of rotating the applied magnetic-field orientation. This work shows how spin misalignment sets a fundamental upper limit for the spin readout fidelity in quantum-computing systems based on PSB.

Designing topological materials with specific topological indices is a complex inverse problem, traditionally tackled through manual, intuition-driven methods that are neither scalable nor efficient for exploring the vast space of possible material configurations. In this work, we develop an algorithm that leverages the covariance matrix adaptation evolution strategy to optimize the Fourier representation of the periodic functions shaping the designer material's characteristics. This includes mass profiles or dielectric tensors for phononic and photonic crystals, respectively, as much as synthetic potentials applicable to electronic and ultra-cold atomic systems. We demonstrate our methodology with a detailed characterization of a class of topological bands known as "fragile topological", showcasing the algorithm's capability to address both topological characteristics and spectral quality. This automation not only streamlines the design process but also significantly expands the potential for identifying and constructing high quality designer topological materials across the wide range of platforms, and is readily extendable to other setups, including higher-dimensional and non-linear systems., Comment: 4 pages + 45 pages SI, 4 + 49 figures

In recent years, serverless computing, especially Function as a Service (FaaS), is rapidly growing in popularity as a cloud programming model. The serverless computing model provides an intuitive interface for developing cloud-based applications, where the development and deployment of scalable microservices has become easier and cost-effective. An increasing number of batch-processing applications are deployed as pipelines that comprise a sequence of functions that must meet their deadline targets to be practical. In this paper, we present our Hybrid Cloud Scheduler (HCS) for orchestrating the execution of serverless batch-processing pipelines deployed over heterogeneous infrastructures. Our framework enables developers to (i) automatically schedule and execute batch-processing applications in heterogeneous environments such as the private edge and public cloud serverless infrastructures, (ii) benefit from cost reduction through the utilization of their own resources in a private cluster, and (iii) significantly improves the probability of meeting the deadline requirements of their applications. Our experimental evaluation demonstrates the efficiency and benefits of our approach.

Serverless computing, also referred to as Function-as-a-Service (FaaS), is a cloud computing model that has attracted significant attention and has been widely adopted in recent years. The serverless computing model offers an intuitive, event-based interface that makes the development and deployment of scalable cloud-based applications easier and cost-effective. An important aspect that has not been examined in these systems is their energy consumption during the application execution. One way to deal with this issue is to schedule the function invocations in an energy-efficient way. However, efficient scheduling of applications in a multi-tenant environment, like FaaS systems, poses significant challenges. The trade-off between the server's energy usage and the hosted functions' performance requirements needs to be taken into consideration. In this work, we propose an Energy Efficient Scheduler for orchestrating the execution of serverless functions so that it minimizes energy consumption while it satisfies the applications' performance demands. Our approach considers real-time performance measurements and historical data and applies a novel DVFS technique to minimize energy consumption. Our detailed experimental evaluation using realistic workloads on our local cluster illustrates the working and benefits of our approach.

Autonomous driving and advanced driver assistance systems (ADAS) rely on cameras to control the driving. In many prior approaches an attacker aiming to stop the vehicle had to send messages on the specialized and better-defended CAN bus. We suggest an easier alternative: manipulate the IP-based network communication between the camera and the ADAS logic, inject fake images of stop signs or red lights into the video stream, and let the ADAS stop the car safely. We created an attack tool that successfully exploits the GigE Vision protocol. Then we analyze two classes of passive anomaly detectors to identify such attacks: protocol-based detectors and video-based detectors. We implemented multiple detectors of both classes and evaluated them on data collected from our test vehicle and also on data from the public BDD corpus. Our results show that such detectors are effective against naive adversaries, but sophisticated adversaries can evade detection. Finally, we propose a novel class of active defense mechanisms that randomly adjust camera parameters during the video transmission, and verify that the received images obey the requested adjustments. Within this class we focus on a specific implementation, the width-varying defense, which randomly modifies the width of every frame. Beyond its function as an anomaly detector, this defense is also a protective measure against certain attacks: by distorting injected image patches it prevents their recognition by the ADAS logic. We demonstrate the effectiveness of the width-varying defense through theoretical analysis and by an extensive evaluation of several types of attack in a wide range of realistic road driving conditions. The best the attack was able to achieve against this defense was injecting a stop sign for a duration of 0.2 seconds, with a success probability of 0.2%, whereas stopping a vehicle requires about 2.5 seconds.

We reframe scene flow as the task of estimating a continuous space-time ODE that describes motion for an entire observation sequence, represented with a neural prior. Our method, EulerFlow, optimizes this neural prior estimate against several multi-observation reconstruction objectives, enabling high quality scene flow estimation via pure self-supervision on real-world data. EulerFlow works out-of-the-box without tuning across multiple domains, including large-scale autonomous driving scenes and dynamic tabletop settings. Remarkably, EulerFlow produces high quality flow estimates on small, fast moving objects like birds and tennis balls, and exhibits emergent 3D point tracking behavior by solving its estimated ODE over long-time horizons. On the Argoverse 2 2024 Scene Flow Challenge, EulerFlow outperforms all prior art, surpassing the next-best unsupervised method by more than 2.5x, and even exceeding the next-best supervised method by over 10%., Comment: Project page at https://vedder.io/eulerflow

This research presents a comprehensive framework for analyzing liquidity in financial markets, particularly in the context of high-frequency trading. By leveraging advanced machine learning classification techniques, including Logistic Regression, Support Vector Machine, and Random Forest, the study aims to predict minute-level price movements using an extensive set of liquidity metrics derived from the Trade and Quote (TAQ) data. The findings reveal that employing a broad spectrum of liquidity measures yields higher predictive accuracy compared to models utilizing a reduced subset of features. Key liquidity metrics, such as Liquidity Ratio, Flow Ratio, and Turnover, consistently emerged as significant predictors across all models, with the Random Forest algorithm demonstrating superior accuracy. This study not only underscores the critical role of liquidity in market stability and transaction costs but also highlights the complexities involved in short-interval market predictions. The research suggests that a comprehensive set of liquidity measures is essential for accurate prediction, and proposes future work to validate these findings across different stock datasets to assess their generalizability.

Drones provide a versatile platform for remote sensing and atmospheric studies. However, strict payload mass limits and intense vibrations have proven obstacles to adoption for astronomy. We present a concept for system-level testing of a long-baseline CubeSat space interferometer using drones, taking advantage of their cm-level xyz station-keeping, 6-dof freedom of movement, large operational environment, access to guide stars for end-to-end testing of optical train and control algorithms, and comparable mass and power requirements. We have purchased two different drone platforms (Aurelia X6 Pro, Freefly Alta X) and present characterization studies of vibrations, flight stability, gps positioning precision, and more. We also describe our progress in sub-system development, including inter-drone laser metrology, realtime gimbal control, and LED beacon tracking. Lastly, we explore whether custom-built drone-borne telescopes could be used for interferometry of bright objects over km-level baselines using vibration-isolation platforms and a small fast delay for fringe-tracking., Comment: submitted to SPIE 2024 (Yokohama)

A quantum system interacting with a time-periodic excitation creates a ladder of hybrid eigenstates in which the system is mixed with an increasing number of photons. This mechanism, referred to as dressing, has been observed in the context of light-matter interaction in systems as varied as atoms, molecules and solid-state qubits. In this work, we demonstrate state dressing in a semiconductor quantum dot tunnel-coupled to a charge reservoir. We observe the emergence of a Floquet ladder of states in the system's high-frequency electrical response, manifesting as interference fringes at the multiphoton resonances despite the system lacking an avoided crossing. We study the dressed quantum dot while changing reservoir temperature, charge lifetime, and excitation amplitude and reveal the fundamental nature of the mechanism by developing a theory based on the quantum dynamics of the Floquet ladder, which is in excellent agreement with the data. Furthermore, we show how the technique finds applications in the accurate electrostatic characterisation of semiconductor quantum dots., Comment: 8 Pages, 4 Figures. Supplementary: 4 Pages, 2 Figures

Motion planning is crucial for safe navigation in complex urban environments. Historically, motion planners (MPs) have been evaluated with procedurally-generated simulators like CARLA. However, such synthetic benchmarks do not capture real-world multi-agent interactions. nuPlan, a recently released MP benchmark, addresses this limitation by augmenting real-world driving logs with closed-loop simulation logic, effectively turning the fixed dataset into a reactive simulator. We analyze the characteristics of nuPlan's recorded logs and find that each city has its own unique driving behaviors, suggesting that robust planners must adapt to different environments. We learn to model such unique behaviors with BehaviorNet, a graph convolutional neural network (GCNN) that predicts reactive agent behaviors using features derived from recently-observed agent histories; intuitively, some aggressive agents may tailgate lead vehicles, while others may not. To model such phenomena, BehaviorNet predicts the parameters of an agent's motion controller rather than directly predicting its spacetime trajectory (as most forecasters do). Finally, we present AdaptiveDriver, a model-predictive control (MPC) based planner that unrolls different world models conditioned on BehaviorNet's predictions. Our extensive experiments demonstrate that AdaptiveDriver achieves state-of-the-art results on the nuPlan closed-loop planning benchmark, improving over prior work by 2% on Test-14 Hard R-CLS, and generalizes even when evaluated on never-before-seen cities., Comment: Project Page: https://arunbalajeev.github.io/world_models_planning/world_model_paper.html

State-of-the-art 3D object detectors are often trained on massive labeled datasets. However, annotating 3D bounding boxes remains prohibitively expensive and time-consuming, particularly for LiDAR. Instead, recent works demonstrate that self-supervised pre-training with unlabeled data can improve detection accuracy with limited labels. Contemporary methods adapt best-practices for self-supervised learning from the image domain to point clouds (such as contrastive learning). However, publicly available 3D datasets are considerably smaller and less diverse than those used for image-based self-supervised learning, limiting their effectiveness. We do note, however, that such 3D data is naturally collected in a multimodal fashion, often paired with images. Rather than pre-training with only self-supervised objectives, we argue that it is better to bootstrap point cloud representations using image-based foundation models trained on internet-scale data. Specifically, we propose a shelf-supervised approach (e.g. supervised with off-the-shelf image foundation models) for generating zero-shot 3D bounding boxes from paired RGB and LiDAR data. Pre-training 3D detectors with such pseudo-labels yields significantly better semi-supervised detection accuracy than prior self-supervised pretext tasks. Importantly, we show that image-based shelf-supervision is helpful for training LiDAR-only, RGB-only and multi-modal (RGB + LiDAR) detectors. We demonstrate the effectiveness of our approach on nuScenes and WOD, significantly improving over prior work in limited data settings. Our code is available at https://github.com/meharkhurana03/cm3d, Comment: The first two authors contributed equally. This work has been accepted to the Conference on Robot Learning (CoRL) 2024

The SWMR atomic register is a fundamental building block in shared memory distributed systems and implementing it from SWSR atomic registers is an important problem. While this problem has been solved in crash-prone systems, it has received less attention in Byzantine systems. Recently, Hu and Toueg gave such an implementation of the SWMR register from SWSR registers. While their definition of register linearizability is consistent with the definition of Byzantine linearizability of a concurrent history of Cohen and Keidar, it has these drawbacks. (1) If the writer is Byzantine, the register is linearizable no matter what values the correct readers return. (2) It ignores values written consistently by a Byzantine writer. We need a stronger notion of a {\em correct write operation}. (3) It allows a value written to just one or a few readers' SWSR registers to be returned, thereby not validating the intention of the writer to write that value honestly. (4) Its notion of a ``current'' value returned by a correct reader is not related to the most recent value written by a correct write operation of a Byzantine writer. We need a more up to date version of the value that can be returned by a correct reader. In this paper, we give a stronger definition of a Byzantine linearizable register that overcomes the above drawbacks. Then we give a construction of a Byzantine linearizable SWMR atomic register from SWSR registers that meets our stronger definition. The construction is correct when $n>3f$, where $n$ is the number of readers, $f$ is the maximum number of Byzantine readers, and the writer can also be Byzantine. The construction relies on a public-key infrastructure., Comment: 18 pages

Extensive research has explored the optical properties of topological insulating materials, driven by their inherent stability and potential applications. In this study, we unveil a novel functionality of three-dimensional integer quantum Hall (3D IQH) states as broad-band filters for circularly polarized light, particularly effective in the terahertz (THz) frequency range under realistic system parameters. We also investigate the impact of practical imperfections, demonstrating the resilience of this filtering effect. Our findings reveal that this phenomenon is independent of the microscopic origin of the 3D IQH state, prompting discussions on its feasibility across diverse candidate materials. These results contribute to our understanding of fundamental optical properties and hold promise for practical applications in optical technologies., Comment: 5 pages, 3 figures

Integrated Speech and Large Language Models (SLMs) that can follow speech instructions and generate relevant text responses have gained popularity lately. However, the safety and robustness of these models remains largely unclear. In this work, we investigate the potential vulnerabilities of such instruction-following speech-language models to adversarial attacks and jailbreaking. Specifically, we design algorithms that can generate adversarial examples to jailbreak SLMs in both white-box and black-box attack settings without human involvement. Additionally, we propose countermeasures to thwart such jailbreaking attacks. Our models, trained on dialog data with speech instructions, achieve state-of-the-art performance on spoken question-answering task, scoring over 80% on both safety and helpfulness metrics. Despite safety guardrails, experiments on jailbreaking demonstrate the vulnerability of SLMs to adversarial perturbations and transfer attacks, with average attack success rates of 90% and 10% respectively when evaluated on a dataset of carefully designed harmful questions spanning 12 different toxic categories. However, we demonstrate that our proposed countermeasures reduce the attack success significantly., Comment: 9+6 pages, Submitted to ACL 2024

Reinforcement Learning is a promising tool for learning complex policies even in fast-moving and object-interactive domains where human teleoperation or hard-coded policies might fail. To effectively reflect this challenging category of tasks, we introduce a dynamic, interactive RL testbed based on robot air hockey. By augmenting air hockey with a large family of tasks ranging from easy tasks like reaching, to challenging ones like pushing a block by hitting it with a puck, as well as goal-based and human-interactive tasks, our testbed allows a varied assessment of RL capabilities. The robot air hockey testbed also supports sim-to-real transfer with three domains: two simulators of increasing fidelity and a real robot system. Using a dataset of demonstration data gathered through two teleoperation systems: a virtualized control environment, and human shadowing, we assess the testbed with behavior cloning, offline RL, and RL from scratch.

Visual control policies can encounter significant performance degradation when visual conditions like lighting or camera position differ from those seen during training -- often exhibiting sharp declines in capability even for minor differences. In this work, we examine robustness to a suite of these types of visual changes for RGB-D and point cloud based visual control policies. To perform these experiments on both model-free and model-based reinforcement learners, we introduce a novel Point Cloud World Model (PCWM) and point cloud based control policies. Our experiments show that policies that explicitly encode point clouds are significantly more robust than their RGB-D counterparts. Further, we find our proposed PCWM significantly outperforms prior works in terms of sample efficiency during training. Taken together, these results suggest reasoning about the 3D scene through point clouds can improve performance, reduce learning time, and increase robustness for robotic learners. Project Webpage: https://pvskand.github.io/projects/PCWM, Comment: Accepted at International Conference on Robotics and Automation, 2024

In this paper, we study the partitioning of a context-aware shared memory data structure so that it can be implemented as a distributed data structure running on multiple machines. By context-aware data structures, we mean that the result of an operation not only depends upon the value of the shared data but also upon the previous operations performed by the same client. While there is substantial work on designing distributed data structures, designing distributed context-aware data structures has not received much attention. We focus on singly-linked lists as a case study of the context-aware data structure. We start with a shared memory context-aware lock-free singly-linked list and show how it can be transformed into a distributed lock-free context-aware singly-linked list. The main challenge in such a transformation is to preserve properties of client-visible operations of the underlying data structure. We present two protocols that preserve these properties of client-visible operations of the linked list. In the first protocol, the distribution is done in the background as a low priority task, while in the second protocol the client-visible operations help the task of distribution without affecting client latency. In both protocols, the client-visible operations remain lock-free. Also, our transformation approach does not utilize any hardware primitives (except a compare-and-swap operation on a single word). We note that our transformation is generic and can be used for other lock-free context-aware data structures that can be constructed from singly-linked lists.

We propose the SAL (Segment Anything in Lidar) method consisting of a text-promptable zero-shot model for segmenting and classifying any object in Lidar, and a pseudo-labeling engine that facilitates model training without manual supervision. While the established paradigm for Lidar Panoptic Segmentation (LPS) relies on manual supervision for a handful of object classes defined a priori, we utilize 2D vision foundation models to generate 3D supervision ``for free''. Our pseudo-labels consist of instance masks and corresponding CLIP tokens, which we lift to Lidar using calibrated multi-modal data. By training our model on these labels, we distill the 2D foundation models into our Lidar SAL model. Even without manual labels, our model achieves $91\%$ in terms of class-agnostic segmentation and $54\%$ in terms of zero-shot Lidar Panoptic Segmentation of the fully supervised state-of-the-art. Furthermore, we outperform several baselines that do not distill but only lift image features to 3D. More importantly, we demonstrate that SAL supports arbitrary class prompts, can be easily extended to new datasets, and shows significant potential to improve with increasing amounts of self-labeled data. Code and models are available at this $\href{https://github.com/nv-dvl/segment-anything-lidar}{URL}$., Comment: Accepted to ECCV 2024

In semiconductor nanostructures, spin blockade (SB) is the most scalable mechanism for electrical spin readout requiring only two bound spins for its implementation which, in conjunction with charge sensing techniques, has led to high-fidelity readout of spins in semiconductor-based quantum processors. However, various mechanisms may lift SB, such as strong spin-orbit coupling (SOC) or low-lying excited states, hence posing challenges to perform spin readout at scale and with high fidelity in such systems. Here, we present a method, based on the dependence of the two-spin system polarizability on energy detuning, to perform spin state readout even when SB lifting mechanisms are dominant. It leverages SB lifting as a resource to detect different spin measurement outcomes selectively and positively. We demonstrate the method using a hybrid system formed by a quantum dot (QD) and a Boron acceptor in a silicon p-type transistor and show spin selective and positive readout of different spin states under SB lifting conditions due to (i) SOC and (ii) low-lying orbital states in the QD. We further use the method to determine the detuning-dependent spin relaxation time of 0.1-8~$\mu$s. Our method should help perform high-fidelity projective spin measurements in systems subject to strong SOC and may facilitate quantum tomography and state leakage studies., Comment: 13 pages, 10 figures

Current scene flow methods broadly fail to describe motion on small objects, and current scene flow evaluation protocols hide this failure by averaging over many points, with most drawn larger objects. To fix this evaluation failure, we propose a new evaluation protocol, Bucket Normalized EPE, which is class-aware and speed-normalized, enabling contextualized error comparisons between object types that move at vastly different speeds. To highlight current method failures, we propose a frustratingly simple supervised scene flow baseline, TrackFlow, built by bolting a high-quality pretrained detector (trained using many class rebalancing techniques) onto a simple tracker, that produces state-of-the-art performance on current standard evaluations and large improvements over prior art on our new evaluation. Our results make it clear that all scene flow evaluations must be class and speed aware, and supervised scene flow methods must address point class imbalances. We release the evaluation code publicly at https://github.com/kylevedder/BucketedSceneFlowEval., Comment: Accepted to ECCV 2024. Project page at https://vedder.io/trackflow

The era of vision-language models (VLMs) trained on web-scale datasets challenges conventional formulations of "open-world" perception. In this work, we revisit the task of few-shot object detection (FSOD) in the context of recent foundational VLMs. First, we point out that zero-shot predictions from VLMs such as GroundingDINO significantly outperform state-of-the-art few-shot detectors (48 vs. 33 AP) on COCO. Despite their strong zero-shot performance, such foundation models may still be sub-optimal. For example, trucks on the web may be defined differently from trucks for a target application such as autonomous vehicle perception. We argue that the task of few-shot recognition can be reformulated as aligning foundation models to target concepts using a few examples. Interestingly, such examples can be multi-modal, using both text and visual cues, mimicking instructions that are often given to human annotators when defining a target concept of interest. Concretely, we propose Foundational FSOD, a new benchmark protocol that evaluates detectors pre-trained on any external data and fine-tuned on multi-modal (text and visual) K-shot examples per target class. We repurpose nuImages for Foundational FSOD, benchmark several popular open-source VLMs, and provide an empirical analysis of state-of-the-art methods. Lastly, we discuss our recent CVPR 2024 Foundational FSOD competition and share insights from the community. Notably, the winning team significantly outperforms our baseline by 23.3 mAP! Our code and dataset splits are available at https://github.com/anishmadan23/foundational_fsod, Comment: The first two authors contributed equally. This work has been accepted to the Neural Information Processing Systems (NeurIPS) 2024 Datasets & Benchmark Track

Contemporary autonomous vehicle (AV) benchmarks have advanced techniques for training 3D detectors, particularly on large-scale multi-modal (LiDAR + RGB) data. Surprisingly, although semantic class labels naturally follow a long-tailed distribution, existing benchmarks only focus on a few common classes (e.g., pedestrian and car) and neglect many rare but crucial classes (e.g., emergency vehicle and stroller). However, AVs must reliably detect both common and rare classes for safe operation in the open world. We address this challenge by formally studying the problem of Long-Tailed 3D Detection (LT3D), which evaluates all annotated classes, including those in-the-tail. We address LT3D with hierarchical losses that promote feature sharing across classes, and introduce diagnostic metrics that award partial credit to ``reasonable'' mistakes with respect to the semantic hierarchy (e.g., mistaking a child for an adult). Further, we point out that rare-class accuracy is particularly improved via multi-modal late fusion (MMLF) of independently trained uni-modal LiDAR and RGB detectors. Importantly, such an MMLF framework allows us to leverage large-scale uni-modal datasets (with more examples for rare classes) to train better uni-modal detectors, unlike prevailing end-to-end trained multi-modal detectors that require paired multi-modal data. Finally, we examine three critical components of our simple MMLF approach from first principles and investigate whether to train 2D or 3D RGB detectors for fusion, whether to match RGB and LiDAR detections in 3D or the projected 2D image plane, and how to fuse matched detections. Our proposed MMLF approach significantly improves LT3D performance over prior work, particularly improving rare class performance from 12.8 to 20.0 mAP!, Comment: The first two authors contributed equally. Project page: https://mayechi.github.io/lt3d-lf-io/