Date: Thu, 9 May 2024 | Total: 345
Event-based semantic segmentation (ESS) is a fundamental yet challenging task for event camera sensing. The difficulties in interpreting and annotating event data limit its scalability. While domain adaptation from images to event data can help to mitigate this issue, there exist data representational differences that require additional effort to resolve. In this work, for the first time, we synergize information from image, text, and event-data domains and introduce OpenESS to enable scalable ESS in an open-world, annotation-efficient manner. We achieve this goal by transferring the semantically rich CLIP knowledge from image-text pairs to event streams. To pursue better cross-modality adaptation, we propose a frame-to-event contrastive distillation and a text-to-event semantic consistency regularization. Experimental results on popular ESS benchmarks showed our approach outperforms existing methods. Notably, we achieve 53.93% and 43.31% mIoU on DDD17 and DSEC-Semantic without using either event or frame labels.
Efficient data utilization is crucial for advancing 3D scene understanding in autonomous driving, where reliance on heavily human-annotated LiDAR point clouds challenges fully supervised methods. Addressing this, our study extends into semi-supervised learning for LiDAR semantic segmentation, leveraging the intrinsic spatial priors of driving scenes and multi-sensor complements to augment the efficacy of unlabeled datasets. We introduce LaserMix++, an evolved framework that integrates laser beam manipulations from disparate LiDAR scans and incorporates LiDAR-camera correspondences to further assist data-efficient learning. Our framework is tailored to enhance 3D scene consistency regularization by incorporating multi-modality, including 1) multi-modal LaserMix operation for fine-grained cross-sensor interactions; 2) camera-to-LiDAR feature distillation that enhances LiDAR feature learning; and 3) language-driven knowledge guidance generating auxiliary supervisions using open-vocabulary models. The versatility of LaserMix++ enables applications across LiDAR representations, establishing it as a universally applicable solution. Our framework is rigorously validated through theoretical analysis and extensive experiments on popular driving perception datasets. Results demonstrate that LaserMix++ markedly outperforms fully supervised alternatives, achieving comparable accuracy with five times fewer annotations and significantly improving the supervised-only baselines. This substantial advancement underscores the potential of semi-supervised approaches in reducing the reliance on extensive labeled data in LiDAR-based 3D scene understanding systems.
Mitigating hallucinations in large vision-language models (LVLMs) remains an open problem. Recent benchmarks do not address hallucinations in open-ended free-form responses, which we term "Type I hallucinations". Instead, they focus on hallucinations responding to very specific question formats -- typically a multiple-choice response regarding a particular object or attribute -- which we term "Type II hallucinations". Additionally, such benchmarks often require external API calls to models which are subject to change. In practice, we observe that a reduction in Type II hallucinations does not lead to a reduction in Type I hallucinations but rather that the two forms of hallucinations are often anti-correlated. To address this, we propose THRONE, a novel object-based automatic framework for quantitatively evaluating Type I hallucinations in LVLM free-form outputs. We use public language models (LMs) to identify hallucinations in LVLM responses and compute informative metrics. By evaluating a large selection of recent LVLMs using public datasets, we show that an improvement in existing metrics do not lead to a reduction in Type I hallucinations, and that established benchmarks for measuring Type I hallucinations are incomplete. Finally, we provide a simple and effective data augmentation method to reduce Type I and Type II hallucinations as a strong baseline.
We introduce a decoder-decoder architecture, YOCO, for large language models, which only caches key-value pairs once. It consists of two components, i.e., a cross-decoder stacked upon a self-decoder. The self-decoder efficiently encodes global key-value (KV) caches that are reused by the cross-decoder via cross-attention. The overall model behaves like a decoder-only Transformer, although YOCO only caches once. The design substantially reduces GPU memory demands, yet retains global attention capability. Additionally, the computation flow enables prefilling to early exit without changing the final output, thereby significantly speeding up the prefill stage. Experimental results demonstrate that YOCO achieves favorable performance compared to Transformer in various settings of scaling up model size and number of training tokens. We also extend YOCO to 1M context length with near-perfect needle retrieval accuracy. The profiling results show that YOCO improves inference memory, prefill latency, and throughput by orders of magnitude across context lengths and model sizes. Code is available at https://aka.ms/YOCO.
Large language models (LLMs) have shown great potential for the automatic generation of feedback in a wide range of computing contexts. However, concerns have been voiced around the privacy and ethical implications of sending student work to proprietary models. This has sparked considerable interest in the use of open source LLMs in education, but the quality of the feedback that such open models can produce remains understudied. This is a concern as providing flawed or misleading generated feedback could be detrimental to student learning. Inspired by recent work that has utilised very powerful LLMs, such as GPT-4, to evaluate the outputs produced by less powerful models, we conduct an automated analysis of the quality of the feedback produced by several open source models using a dataset from an introductory programming course. First, we investigate the viability of employing GPT-4 as an automated evaluator by comparing its evaluations with those of a human expert. We observe that GPT-4 demonstrates a bias toward positively rating feedback while exhibiting moderate agreement with human raters, showcasing its potential as a feedback evaluator. Second, we explore the quality of feedback generated by several leading open-source LLMs by using GPT-4 to evaluate the feedback. We find that some models offer competitive performance with popular proprietary LLMs, such as ChatGPT, indicating opportunities for their responsible use in educational settings.
Diffusion Models (DMs) have exhibited superior performance in generating high-quality and diverse images. However, this exceptional performance comes at the cost of expensive architectural design, particularly due to the attention module heavily used in leading models. Existing works mainly adopt a retraining process to enhance DM efficiency. This is computationally expensive and not very scalable. To this end, we introduce the Attention-driven Training-free Efficient Diffusion Model (AT-EDM) framework that leverages attention maps to perform run-time pruning of redundant tokens, without the need for any retraining. Specifically, for single-denoising-step pruning, we develop a novel ranking algorithm, Generalized Weighted Page Rank (G-WPR), to identify redundant tokens, and a similarity-based recovery method to restore tokens for the convolution operation. In addition, we propose a Denoising-Steps-Aware Pruning (DSAP) approach to adjust the pruning budget across different denoising timesteps for better generation quality. Extensive evaluations show that AT-EDM performs favorably against prior art in terms of efficiency (e.g., 38.8% FLOPs saving and up to 1.53x speed-up over Stable Diffusion XL) while maintaining nearly the same FID and CLIP scores as the full model. Project webpage: https://atedm.github.io.
Powered by large language models (LLMs), AI agents have become capable of many human tasks. Using the most canonical definitions of the Big Five personality, we measure the ability of LLMs to negotiate within a game-theoretical framework, as well as methodological challenges to measuring notions of fairness and risk. Simulations (n=1,500) for both single-issue and multi-issue negotiation reveal increase in domain complexity with asymmetric issue valuations improve agreement rates but decrease surplus from aggressive negotiation. Through gradient-boosted regression and Shapley explainers, we find high openness, conscientiousness, and neuroticism are associated with fair tendencies; low agreeableness and low openness are associated with rational tendencies. Low conscientiousness is associated with high toxicity. These results indicate that LLMs may have built-in guardrails that default to fair behavior, but can be "jail broken" to exploit agreeable opponents. We also offer pragmatic insight in how negotiation bots can be designed, and a framework of assessing negotiation behavior based on game theory and computational social science.
Blockchain technology, while revolutionary in enabling decentralized transactions, faces scalability challenges as the ledger must be replicated across all nodes of the chain, limiting throughput and efficiency. Sharding, which divides the chain into smaller segments, called shards, offers a solution by enabling parallel transaction processing. However, sharding introduces new complexities, notably how to allocate nodes to shards without compromising the network's security. This paper introduces a novel linear optimization framework for node allocation to shards that addresses decentralization constraints while minimizing resource consumption. In contrast to traditional methods that depend on random or trust-based assignments, our approach evaluates node characteristics, including ownership, hardware, and geographical distribution, and requires an explicit specification of decentralization targets with respect to these characteristics. By employing linear optimization, the framework identifies a resource-efficient node set meeting these targets. Adopted by the Internet Computer Protocol (ICP) community, this framework proves its utility in real-world blockchain applications. It provides a quantitative tool for node onboarding and offboarding decisions, balancing decentralization and resource considerations.
Advances in underwater imaging enable the collection of extensive seafloor image datasets that are necessary for monitoring important benthic ecosystems. The ability to collect seafloor imagery has outpaced our capacity to analyze it, hindering expedient mobilization of this crucial environmental information. Recent machine learning approaches provide opportunities to increase the efficiency with which seafloor image datasets are analyzed, yet large and consistent datasets necessary to support development of such approaches are scarce. Here we present BenthicNet: a global compilation of seafloor imagery designed to support the training and evaluation of large-scale image recognition models. An initial set of over 11.4 million images was collected and curated to represent a diversity of seafloor environments using a representative subset of 1.3 million images. These are accompanied by 2.6 million annotations translated to the CATAMI scheme, which span 190,000 of the images. A large deep learning model was trained on this compilation and preliminary results suggest it has utility for automating large and small-scale image analysis tasks. The compilation and model are made openly available for use by the scientific community at https://doi.org/10.20383/103.0614.
This paper proposes a system for chord generation to monophonic symbolic melodies using an LSTM-based model trained on chroma histogram representations of chords. Chroma representations promise more harmonically rich generation than chord label-based approaches, whilst maintaining a small number of dimensions in the dataset. This system is shown to be suitable for limited real-time use. While it does not meet the state-of-the-art for coherent long-term generation, it does show diatonic generation with cadential chord relationships. The need for further study into chroma histograms as an extracted feature in chord generation tasks is highlighted.
The advent of 5G technology promises a paradigm shift in the realm of telecommunications, offering unprecedented speeds and connectivity. However, the efficient management of traffic in 5G networks remains a critical challenge. It is due to the dynamic and heterogeneous nature of network traffic, varying user behaviors, extended network size, and diverse applications, all of which demand highly accurate and adaptable prediction models to optimize network resource allocation and management. This paper investigates the efficacy of live prediction algorithms for forecasting cellular network traffic in real-time scenarios. We apply two live prediction algorithms on machine learning models, one of which is recently proposed Fast LiveStream Prediction (FLSP) algorithm. We examine the performance of these algorithms under two distinct data gathering methodologies: synchronous, where all network cells report statistics simultaneously, and asynchronous, where reporting occurs across consecutive time slots. Our study delves into the impact of these gathering scenarios on the predictive performance of traffic models. Our study reveals that the FLSP algorithm can halve the required bandwidth for asynchronous data reporting compared to conventional online prediction algorithms, while simultaneously enhancing prediction accuracy and reducing processing load. Additionally, we conduct a thorough analysis of algorithmic complexity and memory requirements across various machine learning models. Through empirical evaluation, we provide insights into the trade-offs inherent in different prediction strategies, offering valuable guidance for network optimization and resource allocation in dynamic environments.
The diagnosis and treatment of chest diseases play a crucial role in maintaining human health. X-ray examination has become the most common clinical examination means due to its efficiency and cost-effectiveness. Artificial intelligence analysis methods for chest X-ray images are limited by insufficient annotation data and varying levels of annotation, resulting in weak generalization ability and difficulty in clinical dissemination. Here we present EVA-X, an innovative foundational model based on X-ray images with broad applicability to various chest disease detection tasks. EVA-X is the first X-ray image based self-supervised learning method capable of capturing both semantic and geometric information from unlabeled images for universal X-ray image representation. Through extensive experimentation, EVA-X has demonstrated exceptional performance in chest disease analysis and localization, becoming the first model capable of spanning over 20 different chest diseases and achieving leading results in over 11 different detection tasks in the medical field. Additionally, EVA-X significantly reduces the burden of data annotation in the medical AI field, showcasing strong potential in the domain of few-shot learning. The emergence of EVA-X will greatly propel the development and application of foundational medical models, bringing about revolutionary changes in future medical research and clinical practice. Our codes and models are available at: https://github.com/hustvl/EVA-X.
This paper presents sufficient conditions for the stability and $\ell_2$-gain performance of recurrent neural networks (RNNs) with ReLU activation functions. These conditions are derived by combining Lyapunov/dissipativity theory with Quadratic Constraints (QCs) satisfied by repeated ReLUs. We write a general class of QCs for repeated RELUs using known properties for the scalar ReLU. Our stability and performance condition uses these QCs along with a "lifted" representation for the ReLU RNN. We show that the positive homogeneity property satisfied by a scalar ReLU does not expand the class of QCs for the repeated ReLU. We present examples to demonstrate the stability / performance condition and study the effect of the lifting horizon.
Traffic pattern prediction has emerged as a promising approach for efficiently managing and mitigating the impacts of event-driven bursty traffic in massive machine-type communication (mMTC) networks. However, achieving accurate predictions of bursty traffic remains a non-trivial task due to the inherent randomness of events, and these challenges intensify within live network environments. Consequently, there is a compelling imperative to design a lightweight and agile framework capable of assimilating continuously collected data from the network and accurately forecasting bursty traffic in mMTC networks. This paper addresses these challenges by presenting a machine learning-based framework tailored for forecasting bursty traffic in multi-channel slotted ALOHA networks. The proposed machine learning network comprises long-term short-term memory (LSTM) and a DenseNet with feed-forward neural network (FFNN) layers, where the residual connections enhance the training ability of the machine learning network in capturing complicated patterns. Furthermore, we develop a new low-complexity online prediction algorithm that updates the states of the LSTM network by leveraging frequently collected data from the mMTC network. Simulation results and complexity analysis demonstrate the superiority of our proposed algorithm in terms of both accuracy and complexity, making it well-suited for time-critical live scenarios. We evaluate the performance of the proposed framework in a network with a single base station and thousands of devices organized into groups with distinct traffic-generating characteristics. Comprehensive evaluations and simulations indicate that our proposed machine learning approach achieves a remarkable $52\%$ higher accuracy in long-term predictions compared to traditional methods, without imposing additional processing load on the system.
Graph neural networks (GNNs) are machine learning models specialized for graph data and widely used in many applications. To train GNNs on large graphs that exceed CPU memory, several systems store data on disk and conduct out-of-core processing. However, these systems suffer from either read amplification when reading node features that are usually smaller than a disk page or degraded model accuracy by treating the graph as disconnected partitions. To close this gap, we build a system called DiskGNN, which achieves high I/O efficiency and thus fast training without hurting model accuracy. The key technique used by DiskGNN is offline sampling, which helps decouple graph sampling from model computation. In particular, by conducting graph sampling beforehand, DiskGNN acquires the node features that will be accessed by model computation, and such information is utilized to pack the target node features contiguously on disk to avoid read amplification. Besides, \name{} also adopts designs including four-level feature store to fully utilize the memory hierarchy to cache node features and reduce disk access, batched packing to accelerate the feature packing process, and pipelined training to overlap disk access with other operations. We compare DiskGNN with Ginex and MariusGNN, which are state-of-the-art systems for out-of-core GNN training. The results show that DiskGNN can speed up the baselines by over 8x while matching their best model accuracy.
Objective: We report the development of the patient-centered myAURA application and suite of methods designed to aid epilepsy patients, caregivers, and researchers in making decisions about care and self-management. Materials and Methods: myAURA rests on the federation of an unprecedented collection of heterogeneous data resources relevant to epilepsy, such as biomedical databases, social media, and electronic health records. A generalizable, open-source methodology was developed to compute a multi-layer knowledge graph linking all this heterogeneous data via the terms of a human-centered biomedical dictionary. Results: The power of the approach is first exemplified in the study of the drug-drug interaction phenomenon. Furthermore, we employ a novel network sparsification methodology using the metric backbone of weighted graphs, which reveals the most important edges for inference, recommendation, and visualization, such as pharmacology factors patients discuss on social media. The network sparsification approach also allows us to extract focused digital cohorts from social media whose discourse is more relevant to epilepsy or other biomedical problems. Finally, we present our patient-centered design and pilot-testing of myAURA, including its user interface, based on focus groups and other stakeholder input. Discussion: The ability to search and explore myAURA's heterogeneous data sources via a sparsified multi-layer knowledge graph, as well as the combination of those layers in a single map, are useful features for integrating relevant information for epilepsy. Conclusion: Our stakeholder-driven, scalable approach to integrate traditional and non-traditional data sources, enables biomedical discovery and data-powered patient self-management in epilepsy, and is generalizable to other chronic conditions.
In this work, we present SuFIA, the first framework for natural language-guided augmented dexterity for robotic surgical assistants. SuFIA incorporates the strong reasoning capabilities of large language models (LLMs) with perception modules to implement high-level planning and low-level control of a robot for surgical sub-task execution. This enables a learning-free approach to surgical augmented dexterity without any in-context examples or motion primitives. SuFIA uses a human-in-the-loop paradigm by restoring control to the surgeon in the case of insufficient information, mitigating unexpected errors for mission-critical tasks. We evaluate SuFIA on four surgical sub-tasks in a simulation environment and two sub-tasks on a physical surgical robotic platform in the lab, demonstrating its ability to perform common surgical sub-tasks through supervised autonomous operation under challenging physical and workspace conditions. Project website: orbit-surgical.github.io/sufia
Moderating user-generated content on online platforms is crucial for balancing user safety and freedom of speech. Particularly in the United States, platforms are not subject to legal constraints prescribing permissible content. Each platform has thus developed bespoke content moderation policies, but there is little work towards a comparative understanding of these policies across platforms and topics. This paper presents the first systematic study of these policies from the 43 largest online platforms hosting user-generated content, focusing on policies around copyright infringement, harmful speech, and misleading content. We build a custom web-scraper to obtain policy text and develop a unified annotation scheme to analyze the text for the presence of critical components. We find significant structural and compositional variation in policies across topics and platforms, with some variation attributable to disparate legal groundings. We lay the groundwork for future studies of ever-evolving content moderation policies and their impact on users.
Diffusion models are a powerful generative framework, but come with expensive inference. Existing acceleration methods often compromise image quality or fail under complex conditioning when operating in an extremely low-step regime. In this work, we propose a novel distillation framework tailored to enable high-fidelity, diverse sample generation using just one to three steps. Our approach comprises three key components: (i) Backward Distillation, which mitigates training-inference discrepancies by calibrating the student on its own backward trajectory; (ii) Shifted Reconstruction Loss that dynamically adapts knowledge transfer based on the current time step; and (iii) Noise Correction, an inference-time technique that enhances sample quality by addressing singularities in noise prediction. Through extensive experiments, we demonstrate that our method outperforms existing competitors in quantitative metrics and human evaluations. Remarkably, it achieves performance comparable to the teacher model using only three denoising steps, enabling efficient high-quality generation.
Large Language Models (LLMs) have profoundly changed the world. Their self-attention mechanism is the key to the success of transformers in LLMs. However, the quadratic computational cost $O(n^2)$ to the length $n$ input sequence is the notorious obstacle for further improvement and scalability in the longer context. In this work, we leverage the convolution-like structure of attention matrices to develop an efficient approximation method for attention computation using convolution matrices. We propose a $\mathsf{conv}$ basis system, "similar" to the rank basis, and show that any lower triangular (attention) matrix can always be decomposed as a sum of $k$ structured convolution matrices in this basis system. We then design an algorithm to quickly decompose the attention matrix into $k$ convolution matrices. Thanks to Fast Fourier Transforms (FFT), the attention {\it inference} can be computed in $O(knd \log n)$ time, where $d$ is the hidden dimension. In practice, we have $ d \ll n$, i.e., $d=3,072$ and $n=1,000,000$ for Gemma. Thus, when $kd = n^{o(1)}$, our algorithm achieve almost linear time, i.e., $n^{1+o(1)}$. Furthermore, the attention {\it training forward} and {\it backward gradient} can be computed in $n^{1+o(1)}$ as well. Our approach can avoid explicitly computing the $n \times n$ attention matrix, which may largely alleviate the quadratic computational complexity. Furthermore, our algorithm works on any input matrices. This work provides a new paradigm for accelerating attention computation in transformers to enable their application to longer contexts.
The 3D Human Pose Estimation (3D HPE) task uses 2D images or videos to predict human joint coordinates in 3D space. Despite recent advancements in deep learning-based methods, they mostly ignore the capability of coupling accessible texts and naturally feasible knowledge of humans, missing out on valuable implicit supervision to guide the 3D HPE task. Moreover, previous efforts often study this task from the perspective of the whole human body, neglecting fine-grained guidance hidden in different body parts. To this end, we present a new Fine-Grained Prompt-Driven Denoiser based on a diffusion model for 3D HPE, named \textbf{FinePOSE}. It consists of three core blocks enhancing the reverse process of the diffusion model: (1) Fine-grained Part-aware Prompt learning (FPP) block constructs fine-grained part-aware prompts via coupling accessible texts and naturally feasible knowledge of body parts with learnable prompts to model implicit guidance. (2) Fine-grained Prompt-pose Communication (FPC) block establishes fine-grained communications between learned part-aware prompts and poses to improve the denoising quality. (3) Prompt-driven Timestamp Stylization (PTS) block integrates learned prompt embedding and temporal information related to the noise level to enable adaptive adjustment at each denoising step. Extensive experiments on public single-human pose estimation datasets show that FinePOSE outperforms state-of-the-art methods. We further extend FinePOSE to multi-human pose estimation. Achieving 34.3mm average MPJPE on the EgoHumans dataset demonstrates the potential of FinePOSE to deal with complex multi-human scenarios. Code is available at https://github.com/PKU-ICST-MIPL/FinePOSE_CVPR2024.
Rank and select data structures seek to preprocess a bit vector to quickly answer two kinds of queries: rank(i) gives the number of 1 bits in slots 0 through i, and select(j) gives the first slot s with rank(s) = j. A succinct data structure can answer these queries while using space much smaller than the size of the original bit vector. State of the art succinct rank and select data structures use as little as 4% extra space while answering rank and select queries quickly. Rank queries can be answered using only a handful of array accesses. Select queries can be answered by starting with similar array accesses, followed by a linear scan. Despite these strong results, a tradeoff remains: data structures that use under 4% space are significantly slower at answering rank and select queries than less-space-efficient data structures (using, say, > 20% extra space). In this paper we make significant progress towards closing this gap. We give a new data structure, SPIDER, which uses 3.82% extra space. SPIDER gives the best rank query time for data sets of 8 billion or more bits, even compared to less space-efficient data structures. For select queries, SPIDER outperforms all data structures that use less than 4% space, and significantly closes the gap in select performance between data structures with less than 4% space, and those that use more (over 20%) space. SPIDER makes two main technical contributions. For rank queries, it improves performance by interleaving the metadata with the bit vector to improve cache efficiency. For select queries, it uses predictions to almost eliminate the cost of the linear scan. These predictions are inspired by recent results on data structures with machine-learned predictions, adapted to the succinct data structure setting. Our results hold on both real and synthetic data, showing that these predictions are effective in practice.
We propose a computationally efficient and systematically convergent approach for elastodynamics simulations. We recast the second-order dynamical equation of elastodynamics into an equivalent first-order system of coupled equations, so as to express the solution in the form of a Magnus expansion. With any spatial discretization, it entails computing the exponential of a matrix acting upon a vector. We employ an adaptive Krylov subspace approach to inexpensively and and accurately evaluate the action of the exponential matrix on a vector. In particular, we use an apriori error estimate to predict the optimal Kyrlov subspace size required for each time-step size. We show that the Magnus expansion truncated after its first term provides quadratic and superquadratic convergence in the time-step for nonlinear and linear elastodynamics, respectively. We demonstrate the accuracy and efficiency of the proposed method for one linear (linear cantilever beam) and three nonlinear (nonlinear cantilever beam, soft tissue elastomer, and hyperelastic rubber) benchmark systems. For a desired accuracy in energy, displacement, and velocity, our method allows for $10-100\times$ larger time-steps than conventional time-marching schemes such as Newmark-$\beta$ method. Computationally, it translates to a $\sim$$1000\times$ and $\sim$$10-100\times$ speed-up over conventional time-marching schemes for linear and nonlinear elastodynamics, respectively.
We study the problem of synthesising a two-user broadcast channel using a common message, where each output terminal shares an independent source of randomness with the input terminal. This generalises two problems studied in the literature (Cuff, IEEE Trans. Inform. Theory, 2013; Kurri et.al., IEEE Trans. Inform. Theory, 2021). We give an inner bound on the tradeoff region between the rates of communication and shared randomness, and a lower bound on the minimum communication rate. Although the bounds presented here are not tight in general, they are tight for some special cases, including the aforementioned problems.
Knowing the locations of nearby moving objects is important for a mobile robot to operate safely in a dynamic environment. Dynamic object tracking performance can be improved if robots share observations of tracked objects with nearby team members in real-time. To share observations, a robot must make up-to-date estimates of the transformation from its coordinate frame to the frame of each neighbor, which can be challenging because of odometry drift. We present Multiple Object Tracking with Localization Error Elimination (MOTLEE), a complete system for a multi-robot team to accurately estimate frame transformations and collaboratively track dynamic objects. To accomplish this, robots use open-set image-segmentation methods to build object maps of their environment and then use our Temporally Consistent Alignment of Frames Filter (TCAFF) to align maps and estimate coordinate frame transformations without any initial knowledge of neighboring robot poses. We show that our method for aligning frames enables a team of four robots to collaboratively track six pedestrians with accuracy similar to that of a system with ground truth localization in a challenging hardware demonstration. The code and hardware dataset are available at https://github.com/mit-acl/motlee.