Search

Facing the dramatic increase of real-time applications and time-sensitive services, large-scale ultra-low delay requirements are put forward for the sixth generation (6G) wireless networks. To support massive ultra-reliable and low-latency communications (mURLLC), in this paper we propose an adaptive finite blocklength framework to reduce the over-the-air delay for short packet transmissions with multiple-access and delay-bounded demands. In particular, we first give the specified over-the-air delay model. Then, we reveal the tradeoff relationship among queuing delay, transmission delay, and the number of retransmissions along with the change of finite blocklength, as well as formulate the adaptive blocklength framework. Based on the adaptive blocklength framework and associated with grant-free (GF) access protocol, we formulate the average over-the-air delay minimization problem, where the blocklength can be adaptively changed in terms of transmission time interval (TTI) design and bandwidth allocation to achieve the optimal tradeoff and obtain its minimum over-the-air delay. We develop the cooperative multi-agent deep Q-network (M-DQN) scheme with a grouping mechanism to efficiently solve the average over-the-air delay minimization problem. Numerical results validate our proposed adaptive blocklength scheme outperforms corresponding schemes in long-term evolution (LTE) and the fifth generation (5G) new radio (NR).

As the number of real-time applications with ultra-low delay requirements quickly grows, massive ultra-reliable and low-latency communication (mURLLC) has been proposed to provide a wide range of delay-sensitive services for the sixth generation (6G) wireless networks. However, it is difficult to meet the stringent delay demand of massive connectivity with existing grant-based (GB) random access and fixed frame structure in long-term evolution (LTE) and the fifth generation (5G) new radio (NR) systems. To solve this problem, in this paper we propose the new grant-free (GF) based adaptive blocklength scheme for short packet transmission to reduce the access delay. We develop the adaptive blocklength framework where the blocklength can be adaptively changed according to the real-time load, to revise the traditional non-flexible frame structure which impacts the delay performance. Taking the features of mURLLC into consideration, we analyze the GF random access procedure, packet arrival behavior, packet collision, and packet transmission error in the finite blocklength (FB) regime. On this basis, we derive the closed-form expression of successful access and transmission probability and give the GF-based status update model. Then, we propose the access delay minimization problem that jointly considers queuing delay and transmission delay to reduce the overall access delay. With the alternating optimization algorithm, we obtain the optimal blocklength of each packet, thus forming the corresponding adaptive blocklength scheme for mURLLC. Simulation results verify the correctness of theoretical results and show that our proposed adaptive blocklength scheme can significantly reduce the access delay compared with that of LTE and 5G NR systems.

Rate splitting multiple access (RSMA) is one of the most promising techniques for ultra-reliable and low-latency communications (URLLC) with stringent requirements on delay and reliability of multiple access. To fully explore the delay performance enhancement brought by uplink RSMA to URLLC, in this paper, we evaluate the performance of two-user uplink RSMA and propose the corresponding blocklength minimization problem. We analyze the impact of finite blocklength (FBL) code on the achievable rate region and the effective throughput of uplink RSMA. On this basis, we propose the problem of minimizing the blocklength for uplink RSMA with power allocation under constrained reliability and effective throughput. Then, we present an alternating optimization method to solve this non-convex problem. Simulation results show that different from the infinite blocklength (IBL) regime, the achievable rate region of the uplink RSMA is not always larger than that of uplink non-orthogonal multiple access (NOMA) in the FBL regime. But with the help of our proposed blocklength minimization scheme, uplink RSMA can achieve the same achievable rate with a smaller blocklength compared to uplink NOMA, frequency division multiple access (FDMA), and time division multiple access (TDMA) without the need for time sharing in the FBL regime, showing the potential of uplink RSMA to achieve low delay for URLLC.

Energy efficiency (EE) is one of the most important metrics for the beyond fifth generation (B5G) and the future sixth generation (6G) wireless networks. Reconfigurable intelligent surface (RIS) has been widely focused on EE enhancement for wireless networks because it is power-saving, programmable, and easy to be deployed. However, RIS is generally passive and thus difficult to obtain corresponding full channel state information (CSI), which severely impacts the EE enhancement of RIS-assisted wireless communications. To solve this problem, in this paper we propose the new single-active-antenna combined RIS transmitter structure, which can replace traditional multiple antennas to reduce hardware cost and power consumption. Based on the single-active-antenna combined RIS structure, we develop the Dumb RIS-Assisted Random Beamforming (Darb)-based Joint RIS-Elements and Transmit-power optimizAtion (Jeta) scheme, where dumb RIS randomly changes its phase shift according to isotropic distribution only depending on the CSI feedback from users to RIS-assisted transmitter. Then, we jointly design the number of RIS elements and optimize the transmit power to maximize the EE of RIS-assisted wireless communications. Simulation results show that compared with the traditional multi-antenna system, our developed Darb-based-Jeta scheme can significantly increase the EE without the full CSI., Comment: 6 pages, 4 figures

With various time-sensitive tasks to be served, ultra-reliable and low-latency communications (URLLC) has become one of the most important scenarios for the fifth generation (5G) wireless communications. The end-to-end delay from the sub-millisecond-level to the second-level is first put forward for a wide range of delay-sensitive tasks in the future sixth generation (6G) communication networks, which imposes a strict requirement on satisfying various real-time services and applications with different stringent quality of service (QoS) demands. Thus, we need to find out new delay reduction framework to satisfy the more stringent delay requirements. In this article, a state-of-the-art overview of end-to-end delay composition and delay analysis combined with access protocols are elaborated. We reveal the tradeoff relationship among transmission delay, queuing delay, and retransmission times with the change of blocklength in the finite blocklength (FBL) regime. Based on the tradeoff and combining the grant-free (GF) random access (RA) scheme, we propose the adaptive blocklength framework and investigate several effective algorithms for efficiently reducing the over-the-air delay. Numerical results show that our proposed framework and schemes can significantly reduce the over-the-air delay for URLLC., Comment: 7 pages, 5 figures

Recently, unmanned aerial vehicle (UAV) has attracted much attention due to its flexible deployment and controllable mobility. As the general communication network cannot meet the emergency requirements, in this paper we study the multi-UAV enabled wireless emergency communication system. Our goal is to maximize the capacity with jointly optimizing trajectory and allocating power. To tackle this non-convex optimization problem, we first decompose it into two sub-problems to optimize the trajectory and power allocation, respectively. Then, we propose the successive convex approximation technique and the block coordinate update algorithm to solve the two subproblems. The approximate optimal solution can be obtained after continuous iterations. Simulation results show that the capacity can be greatly increased using our proposed joint trajectory optimization and power allocation., Comment: 6 pages, 3 figures

Food recommendation systems serve as pivotal components in the realm of digital lifestyle services, designed to assist users in discovering recipes and food items that resonate with their unique dietary predilections. Typically, multi-modal descriptions offer an exhaustive profile for each recipe, thereby ensuring recommendations that are both personalized and accurate. Our preliminary investigation of two datasets indicates that pre-trained multi-modal dense representations might precipitate a deterioration in performance compared to ID features when encapsulating interactive relationships. This observation implies that ID features possess a relative superiority in modeling interactive collaborative signals. Consequently, contemporary cutting-edge methodologies augment ID features with multi-modal information as supplementary features, overlooking the latent semantic relations between recipes. To rectify this, we present CLUSSL, a novel food recommendation framework that employs clustering and self-supervised learning. Specifically, CLUSSL formulates a modality-specific graph tailored to each modality with discrete/continuous features, thereby transforming semantic features into structural representation. Furthermore, CLUSSL procures recipe representations pertinent to different modalities via graph convolutional operations. A self-supervised learning objective is proposed to foster independence between recipe representations derived from different unimodal graphs. Comprehensive experiments on real-world datasets substantiate that CLUSSL consistently surpasses state-of-the-art recommendation benchmarks in performance., Comment: Working paper

LLMs have achieved significant performance progress in various NLP applications. However, LLMs still struggle to meet the strict requirements for accuracy and reliability in the medical field and face many challenges in clinical applications. Existing clinical diagnostic evaluation benchmarks for evaluating medical agents powered by LLMs have severe limitations. Firstly, most existing medical evaluation benchmarks face the risk of data leakage or contamination. Secondly, existing benchmarks often neglect the characteristics of multiple departments and specializations in modern medical practice. Thirdly, existing evaluation methods are limited to multiple-choice questions, which do not align with the real-world diagnostic scenarios. Lastly, existing evaluation methods lack comprehensive evaluations of end-to-end real clinical scenarios. These limitations in benchmarks in turn obstruct advancements of LLMs and agents for medicine. To address these limitations, we introduce ClinicalLab, a comprehensive clinical diagnosis agent alignment suite. ClinicalLab includes ClinicalBench, an end-to-end multi-departmental clinical diagnostic evaluation benchmark for evaluating medical agents and LLMs. ClinicalBench is based on real cases that cover 24 departments and 150 diseases. ClinicalLab also includes four novel metrics (ClinicalMetrics) for evaluating the effectiveness of LLMs in clinical diagnostic tasks. We evaluate 17 LLMs and find that their performance varies significantly across different departments. Based on these findings, in ClinicalLab, we propose ClinicalAgent, an end-to-end clinical agent that aligns with real-world clinical diagnostic practices. We systematically investigate the performance and applicable scenarios of variants of ClinicalAgent on ClinicalBench. Our findings demonstrate the importance of aligning with modern medical practices in designing medical agents.

Topological flat bands formed in two-dimensional lattice systems offer unique opportunity to study the fractional phases of matter in the absence of an external magnetic field. Celebrated examples include fractional quantum anomalous Hall (FQAH) effects and fractional topological insulators. Recently, FQAH effects have been experimentally realized in both the twisted bilayer MoTe2 (tMoTe2) system and the rhombohedral stacked multilayer graphene/hBN moir\'e systems. To date, experimental studies mainly focus on the first moir\'e flat band, except a very recent work that reported the evidence of the integer and fractional quantum spin Hall effects in higher moir\'e bands of a 2.1{\deg} tMoTe2 device. Here, we present the systematical transport study of approximately 3{\deg} tMoTe2 devices, especially for the second moir\'e band. At {\nu} = -2 and -4, time-reversal-symmetric single and double quantum spin Hall states formed, consistent with the previous observation in 2.1{\deg} tMoTe2 device. On the other hand, we observed ferromagnetism in the second moir\'e band, and a Chern insulator state driven by out-of-plane magnetic fields at {\nu} = -3. At {\nu} = -2.5, nonmonotonic temperature dependence of resistivity and large out-of-plane negative magnetoresistance have been observed, which likely arises from the frustrated liquid-like ground state and weak effective Ruderman-Kittel-Kasuya-Yosida interactions. Applying out-of-plane electric field can induce quantum phase transitions at both integer and fractional filling factors. Our studies pave the way for realizing tunable topological states and other unexpected magnetic phases beyond the first moir\'e flat band based on twisted MoTe2 platform.

The dewetting of thin nanofilms is significantly impacted by thermal fluctuations, liquid-solid slip, and disjoining pressure, which can be described by lubrication equations augmented by appropriately scaled noise terms, known as stochastic lubrication equations. Here molecular dynamics simulations along with a newly proposed slip-generating method are adopted to study the instability of nanofilms with arbitrary slip. These simulations show that strong-slip dewetting is distinct from weak-slip dewetting by faster growth of perturbations and fewer droplets after dewetting, which can not be predicted by the existing stochastic lubrication equation. A new stochastic lubrication equation considering the strong slip boundary condition is thus derived using a long-wave approximation to the equations of fluctuating hydrodynamics. The linear stability analysis of this equation, i.e., surface spectrum, agrees well with molecular simulations. Interestingly, strong slip can break down the usual Stokes limits adopted in weak-slip dewetting and bring the inertia into effect. The evolution of the standard deviation of the film height $W^2(t)={\overline{h^2}-{\overline{h}}^2}$ at the initial stage of the strong-slip dewetting is found to be $W\sim t^{1/4}$ in contrast to $W\sim t^{1/8}$ for the weak-slip dewetting.

Understanding the stability mechanism of surface micro/nanobubbles adhered to gas-evolving electrodes is essential for improving the efficiency of water electrolysis, which is known to be hindered by the bubble coverage on electrodes. Using molecular simulations, the diffusion-controlled evolution of single electrolytic nanobubbles on wettability-patterned nanoelectrodes is investigated. These nanoelectrodes feature hydrophobic islands as preferential nucleation sites and allow the growth of nanobubbles in the pinning mode. In these simulations, a threshold current density distinguishing stable nanobubbles from unstable nanobubbles is found. When the current density remains below the threshold value, nucleated nanobubbles grow to their equilibrium states, maintaining their nanoscopic size. However, for the current density above the threshold value, nanobubbles undergo unlimited growth and can eventually detach due to buoyancy. Increasing the pinning length of nanobubbles increases the degree of nanobubble instability. By connecting the current density with the local gas oversaturation, an extension of the stability theory for surface nanobubbles [Lohse and Zhang, Phys. Rev. E, 2015, 91, 031003(R)] accurately predicts the nanobubble behavior found in molecular simulations, including equilibrium contact angles and the threshold current density. For larger systems that are not accessible to molecular simulations, continuum numerical simulations with the finite difference method combined with the immersed boundary method are performed, again demonstrating good agreement between numerics and theories.

Effective slip boundary conditions for flows over periodic micro-structured surfaces containing a secondary immiscible fluid are derived. The primary fluid is in the Cassie state, while the geometries of the micro-structures can be arbitrary. We investigate the impact of the second immiscible fluid on external flow, introducing the effect caused by the viscosity difference between two fluids and the inertia effect of the second fluid into classic Navier slip condition. The effective slip length obtained from our theory for flows over rectangular micro-structures is in good agreement with prior analytical findings. We also apply the theory to mushroom-like micro-structures. The derived effective slip velocity also matches well with two-phase numerical simulations. Implementing the slip boundary conditions on micro-structured surfaces produces external flow fields that are aligned well with the simulation results. Employing a multi-scale homogenization method, we dispose of two-phase flows characterized by strong coupling at the fluid-fluid interface. By introducing the framework of lid-driven cavity approximation, our theory finds practical applications across various scenarios.

Weakly-supervised temporal action localization aims to localize action instances in videos with only video-level action labels. Existing methods mainly embrace a localization-by-classification pipeline that optimizes the snippet-level prediction with a video classification loss. However, this formulation suffers from the discrepancy between classification and detection, resulting in inaccurate separation of foreground and background (F\&B) snippets. To alleviate this problem, we propose to explore the underlying structure among the snippets by resorting to unsupervised snippet clustering, rather than heavily relying on the video classification loss. Specifically, we propose a novel clustering-based F\&B separation algorithm. It comprises two core components: a snippet clustering component that groups the snippets into multiple latent clusters and a cluster classification component that further classifies the cluster as foreground or background. As there are no ground-truth labels to train these two components, we introduce a unified self-labeling mechanism based on optimal transport to produce high-quality pseudo-labels that match several plausible prior distributions. This ensures that the cluster assignments of the snippets can be accurately associated with their F\&B labels, thereby boosting the F\&B separation. We evaluate our method on three benchmarks: THUMOS14, ActivityNet v1.2 and v1.3. Our method achieves promising performance on all three benchmarks while being significantly more lightweight than previous methods. Code is available at https://github.com/Qinying-Liu/CASE, Comment: ICCV2023

Deep neural networks (DNNs) have been deployed for many image segmentation tasks and achieved outstanding performance. However, preparing a dataset for training segmentation DNNs is laborious and costly since typically pixel-level annotations are provided for each object of interest. To alleviate this issue, one can provide only weak labels such as bounding boxes or scribbles, or less accurate (noisy) annotations of the objects. These are significantly faster to generate and thus result in more annotated images given the same time budget. However, the reduction in quality might negatively affect the segmentation performance of the resulting model. In this study, we perform a thorough cost-effectiveness evaluation of several weak and noisy labels. We considered 11 variants of annotation strategies and 4 datasets. We conclude that the common practice of accurately outlining the objects of interest is virtually never the optimal approach when the annotation time is limited, even if notable annotation time is available (10s of hours). Annotation approaches that stood out in such scenarios were (1) contour-based annotation with rough continuous traces, (2) polygon-based annotation with few vertices, and (3) box annotations combined with the Segment Anything Model (SAM). In situations where unlimited annotation time was available, precise annotations still lead to the highest segmentation model performance.

In this article, we study the collocated and distributed deployment of intelligent reflecting surfaces (IRS) for a fixed total number of IRS elements to support enhanced mobile broadband (eMBB) and ultra-reliable low-latency communication (URLLC) services inside a factory. We build a channel model that incorporates the line-of-sight (LOS) probability and power loss of each transmission path, and propose three metrics, namely, the expected received signal-to-noise ratio (SNR), expected finite-blocklength (FB) capacity, and expected outage probability, where the expectation is taken over the probability distributions of interior blockages and channel fading. The expected received SNR and expected FB capacity for extremely high blockage densities are derived in closed-form as functions of the amount and height of IRSs and the density, size, and penetration loss of blockages, which are verified by Monte Carlo simulations. Results show that deploying IRSs vertically higher leads to higher expected received SNR and expected FB capacity. By analysing the average/minimum/maximum of the three metrics versus the number of IRSs, we find that for high blockage densities, both eMBB and URLLC services benefit from distributed deployment; and for low blockage densities, URLLC services benefit from distributed deployment while eMBB services see limited difference between collocated and distributed deployment.

Computer simulations offer a robust toolset for exploring complex systems across various disciplines. A particularly impactful approach within this realm is Agent-Based Modeling (ABM), which harnesses the interactions of individual agents to emulate intricate system dynamics. ABM's strength lies in its bottom-up methodology, illuminating emergent phenomena by modeling the behaviors of individual components of a system. Yet, ABM has its own set of challenges, notably its struggle with modeling natural language instructions and common sense in mathematical equations or rules. This paper seeks to transcend these boundaries by integrating Large Language Models (LLMs) like GPT into ABM. This amalgamation gives birth to a novel framework, Smart Agent-Based Modeling (SABM). Building upon the concept of smart agents -- entities characterized by their intelligence, adaptability, and computation ability -- we explore in the direction of utilizing LLM-powered agents to simulate real-world scenarios with increased nuance and realism. In this comprehensive exploration, we elucidate the state of the art of ABM, introduce SABM's potential and methodology, and present three case studies (source codes available at https://github.com/Roihn/SABM), demonstrating the SABM methodology and validating its effectiveness in modeling real-world systems. Furthermore, we cast a vision towards several aspects of the future of SABM, anticipating a broader horizon for its applications. Through this endeavor, we aspire to redefine the boundaries of computer simulations, enabling a more profound understanding of complex systems., Comment: Source codes are available at https://github.com/Roihn/SABM

Instances of casualties resulting from large crowds persist, highlighting the existing limitations of current crowd management practices in Smart Cities. One notable drawback is the insufficient provision for disadvantaged individuals who may require additional time to evacuate due to their slower running speed. Moreover, the existing escape strategies may fall short of ensuring the safety of all individuals during a crowd surge. To address these pressing concerns, this paper proposes two crowd management methodologies. Firstly, we advocate for implementing a fair evacuation strategy following a surge event, which considers the diverse needs of all individuals, ensuring inclusivity and mitigating potential risks. Secondly, we propose a preventative approach involving the adjustment of attraction locations and switching between stage performances in large-crowded events to minimize the occurrence of surges and enhance crowd dispersion. We used high-fidelity crowd management simulators to assess the effectiveness of our proposals. Our findings demonstrate the positive impact of the fair evacuation strategy on safety measures and inclusivity, which increases fairness by 41.8% on average. Furthermore, adjusting attraction locations and stage performances has shown a significant reduction in surges by 34% on average, enhancing overall crowd safety.

Source-free object detection (SFOD) aims to adapt a source-trained detector to an unlabeled target domain without access to the labeled source data. Current SFOD methods utilize a threshold-based pseudo-label approach in the adaptation phase, which is typically limited to high-confidence pseudo-labels and results in a loss of information. To address this issue, we propose a new approach to take full advantage of pseudo-labels by introducing high and low confidence thresholds. Specifically, the pseudo-labels with confidence scores above the high threshold are used conventionally, while those between the low and high thresholds are exploited using the Low-confidence Pseudo-labels Utilization (LPU) module. The LPU module consists of Proposal Soft Training (PST) and Local Spatial Contrastive Learning (LSCL). PST generates soft labels of proposals for soft training, which can mitigate the label mismatch problem. LSCL exploits the local spatial relationship of proposals to improve the model's ability to differentiate between spatially adjacent proposals, thereby optimizing representational features further. Combining the two components overcomes the challenges faced by traditional methods in utilizing low-confidence pseudo-labels. Extensive experiments on five cross-domain object detection benchmarks demonstrate that our proposed method outperforms the previous SFOD methods, achieving state-of-the-art performance.

Conventional imaging systems can only capture light intensity. Meanwhile, the lost phase information may be critical for a variety of applications such as label-free microscopy and optical metrology. Existing phase retrieval techniques typically require a bulky setup, multi-frame measurements, or prior information of the target scene. Here, we proposed an extremely compact system for complex amplitude imaging, leveraging the extreme versatility of a single-layer metalens to generate spatially-multiplexed and polarization-phase-shifted point spread functions. Combining the metalens with a polarization camera, the system can simultaneously record four polarization shearing interference patterns along both in-plane directions, thus allowing the deterministic reconstruction of the complex amplitude light field in a single shot. Using an incoherent light-emitting diode as the illumination, we experimentally demonstrated speckle-noise-free complex amplitude imaging for both static and moving objects with tailored magnification ratio and field-of-view. The miniaturized and robust system may open the door for complex amplitude imaging in portable devices for point-of-care applications.

Thermal motions of molecules can generate nanowaves on the free surface of a liquid film. As nanofilms are susceptible to the contamination of surfactants, this work investigates the effects of surfactants on dynamics of nanowaves on a bounded film with a finite depth, using both molecular dynamics simulations and analytical theories. In molecular simulations, a bead-spring model is adopted to simulate surfactants, where beads are connected by the finite extensive nonlinear elastic potentials. Fourier transforms of the film surface profiles $h(x,t)$ extracted from molecular simulations are performed to obtain the static spectrum $|h_q|_{\mathrm{rms}}$ and temporal correlations of surface modes $$. It is shown that the spectral amplitude is increased for the contaminated liquid surface compared to the clean surface because surfactants can decrease surface tension. A higher concentration of surfactants on the surface not only decreases the surface tension but also causes elastic energy to the free surface, as the scaling of spectral amplitude with wavenumbers changes from $|h_q|_{\mathrm{rms}}\sim q^{-1}$ to $|h_q|_{\mathrm{rms}}\sim q^{-2}$ for modes with large wavenumbers. Regarding the temporal correlations of surface modes, it is observed that the presence of surfactants leads to a slower decay, which, however, cannot be predicted by only considering the decreased surface tension. Based on the Boussinesq Scriven model for surface viscosity, a linear stability analysis of Stokes flow for films with arbitrary depth is conducted and the obtained dispersion relation considering surface viscosity can justify the simulation results.

The efficiency of water electrolysis is significantly impacted by the generation of micro- and nanobubbles on the electrodes. Here molecular dynamics simulations are used to investigate the dynamics of single electrolytic nanobubbles on nanoelectrodes. The simulations reveal that, depending on the value of current, nucleated nanobubbles either grow to an equilibrium state or grow unlimitedly and then detach. To account for these findings, the stability theory for surface nanobubbles is generalized by incorporating the electrolytic gas influx at the nanobubble's contact line and adopting a real gas law, leading to accurate predictions for the numerically observed transient growth and stationary states of the nanobubbles. With this theory, the minimum current for bubble detachment can also be analytically derived. In the detachment regime, the radius of the nanobubble first increases as $R\propto t^{1/2}$ and then as $R\propto t^{1/3}$, up to bubble detachment.

Shape learning, or the ability to leverage shape information, could be a desirable property of convolutional neural networks (CNNs) when target objects have specific shapes. While some research on the topic is emerging, there is no systematic study to conclusively determine whether and under what circumstances CNNs learn shape. Here, we present such a study in the context of segmentation networks where shapes are particularly important. We define shape and propose a new behavioral metric to measure the extent to which a CNN utilizes shape information. We then execute a set of experiments with synthetic and real-world data to progressively uncover under which circumstances CNNs learn shape and what can be done to encourage such behavior. We conclude that (i) CNNs do not learn shape in typical settings but rather rely on other features available to identify the objects of interest, (ii) CNNs can learn shape, but only if the shape is the only feature available to identify the object, (iii) sufficiently large receptive field size relative to the size of target objects is necessary for shape learning; (iv) a limited set of augmentations can encourage shape learning; (v) learning shape is indeed useful in the presence of out-of-distribution data., Comment: Accepted by Pattern Recognition, reformatted for Arxiv for readability

Training segmentation models for medical images continues to be challenging due to the limited availability of data annotations. Segment Anything Model (SAM) is a foundation model that is intended to segment user-defined objects of interest in an interactive manner. While the performance on natural images is impressive, medical image domains pose their own set of challenges. Here, we perform an extensive evaluation of SAM's ability to segment medical images on a collection of 19 medical imaging datasets from various modalities and anatomies. We report the following findings: (1) SAM's performance based on single prompts highly varies depending on the dataset and the task, from IoU=0.1135 for spine MRI to IoU=0.8650 for hip X-ray. (2) Segmentation performance appears to be better for well-circumscribed objects with prompts with less ambiguity and poorer in various other scenarios such as the segmentation of brain tumors. (3) SAM performs notably better with box prompts than with point prompts. (4) SAM outperforms similar methods RITM, SimpleClick, and FocalClick in almost all single-point prompt settings. (5) When multiple-point prompts are provided iteratively, SAM's performance generally improves only slightly while other methods' performance improves to the level that surpasses SAM's point-based performance. We also provide several illustrations for SAM's performance on all tested datasets, iterative segmentation, and SAM's behavior given prompt ambiguity. We conclude that SAM shows impressive zero-shot segmentation performance for certain medical imaging datasets, but moderate to poor performance for others. SAM has the potential to make a significant impact in automated medical image segmentation in medical imaging, but appropriate care needs to be applied when using it., Comment: Link to our code: https://github.com/mazurowski-lab/segment-anything-medical-evaluation