Search

Enhancing and preserving the readability of document images, particularly historical ones, is crucial for effective document image analysis. Numerous models have been proposed for this task, including convolutional-based, transformer-based, and hybrid convolutional-transformer architectures. While hybrid models address the limitations of purely convolutional or transformer-based methods, they often suffer from issues like quadratic time complexity. In this work, we propose a Mamba-based architecture for document binarisation, which efficiently handles long sequences by scaling linearly and optimizing memory usage. Additionally, we introduce novel modifications to the skip connections by incorporating Difference of Gaussians (DoG) features, inspired by conventional signal processing techniques. These multiscale high-frequency features enable the model to produce high-quality, detailed outputs.

Image and video generative models that are pre-trained on Internet-scale data can greatly increase the generalization capacity of robot learning systems. These models can function as high-level planners, generating intermediate subgoals for low-level goal-conditioned policies to reach. However, the performance of these systems can be greatly bottlenecked by the interface between generative models and low-level controllers. For example, generative models may predict photorealistic yet physically infeasible frames that confuse low-level policies. Low-level policies may also be sensitive to subtle visual artifacts in generated goal images. This paper addresses these two facets of generalization, providing an interface to effectively "glue together" language-conditioned image or video prediction models with low-level goal-conditioned policies. Our method, Generative Hierarchical Imitation Learning-Glue (GHIL-Glue), filters out subgoals that do not lead to task progress and improves the robustness of goal-conditioned policies to generated subgoals with harmful visual artifacts. We find in extensive experiments in both simulated and real environments that GHIL-Glue achieves a 25% improvement across several hierarchical models that leverage generative subgoals, achieving a new state-of-the-art on the CALVIN simulation benchmark for policies using observations from a single RGB camera. GHIL-Glue also outperforms other generalist robot policies across 3/4 language-conditioned manipulation tasks testing zero-shot generalization in physical experiments., Comment: Code, model checkpoints and videos can be found at https://ghil-glue.github.io

6D object pose estimation involves determining the three-dimensional translation and rotation of an object within a scene and relative to a chosen coordinate system. This problem is of particular interest for many practical applications in industrial tasks such as quality control, bin picking, and robotic manipulation, where both speed and accuracy are critical for real-world deployment. Current models, both classical and deep-learning-based, often struggle with the trade-off between accuracy and latency. Our research focuses on enhancing the speed of a prominent state-of-the-art deep learning model, GDRNPP, while keeping its high accuracy. We employ several techniques to reduce the model size and improve inference time. These techniques include using smaller and quicker backbones, pruning unnecessary parameters, and distillation to transfer knowledge from a large, high-performing model to a smaller, more efficient student model. Our findings demonstrate that the proposed configuration maintains accuracy comparable to the state-of-the-art while significantly improving inference time. This advancement could lead to more efficient and practical applications in various industrial scenarios, thereby enhancing the overall applicability of 6D Object Pose Estimation models in real-world settings.

Advances in microscopy imaging enable researchers to visualize structures at the nanoscale level thereby unraveling intricate details of biological organization. However, challenges such as image noise, photobleaching of fluorophores, and low tolerability of biological samples to high light doses remain, restricting temporal resolutions and experiment durations. Reduced laser doses enable longer measurements at the cost of lower resolution and increased noise, which hinders accurate downstream analyses. Here we train a denoising diffusion probabilistic model (DDPM) to predict high-resolution images by conditioning the model on low-resolution information. Additionally, the probabilistic aspect of the DDPM allows for repeated generation of images that tend to further increase the signal-to-noise ratio. We show that our model achieves a performance that is better or similar to the previously best-performing methods, across four highly diverse datasets. Importantly, while any of the previous methods show competitive performance for some, but not all datasets, our method consistently achieves high performance across all four data sets, suggesting high generalizability.

We study the capability of Video-Language (VidL) models in understanding compositions between objects, attributes, actions and their relations. Composition understanding becomes particularly challenging for video data since the compositional relations rapidly change over time in videos. We first build a benchmark named AARO to evaluate composition understanding related to actions on top of spatial concepts. The benchmark is constructed by generating negative texts with incorrect action descriptions for a given video and the model is expected to pair a positive text with its corresponding video. Furthermore, we propose a training method called NAVERO which utilizes video-text data augmented with negative texts to enhance composition understanding. We also develop a negative-augmented visual-language matching loss which is used explicitly to benefit from the generated negative text. We compare NAVERO with other state-of-the-art methods in terms of compositional understanding as well as video-text retrieval performance. NAVERO achieves significant improvement over other methods for both video-language and image-language composition understanding, while maintaining strong performance on traditional text-video retrieval tasks.

Modern artificial intelligence (AI) systems are powered by foundation models. This paper presents a new set of foundation models, called Llama 3. It is a herd of language models that natively support multilinguality, coding, reasoning, and tool usage. Our largest model is a dense Transformer with 405B parameters and a context window of up to 128K tokens. This paper presents an extensive empirical evaluation of Llama 3. We find that Llama 3 delivers comparable quality to leading language models such as GPT-4 on a plethora of tasks. We publicly release Llama 3, including pre-trained and post-trained versions of the 405B parameter language model and our Llama Guard 3 model for input and output safety. The paper also presents the results of experiments in which we integrate image, video, and speech capabilities into Llama 3 via a compositional approach. We observe this approach performs competitively with the state-of-the-art on image, video, and speech recognition tasks. The resulting models are not yet being broadly released as they are still under development.

MRI is a widely used ionization-free soft-tissue imaging modality, often employed repeatedly over a patient's lifetime. However, prolonged scanning durations, among other issues, can limit availability and accessibility. In this work, we aim to substantially reduce scan times by leveraging prior scans of the same patient. These prior scans typically contain considerable shared information with the current scan, thereby enabling higher acceleration rates when appropriately utilized. We propose a prior informed reconstruction method with a trained diffusion model in conjunction with data-consistency steps. Our method can be trained with unlabeled image data, eliminating the need for a dataset of either k-space measurements or paired longitudinal scans as is required of other learning-based methods. We demonstrate superiority of our method over previously suggested approaches in effectively utilizing prior information without over-biasing prior consistency, which we validate on both an open-source dataset of healthy patients as well as several longitudinal cases of clinical interest.

This study presents a novel framework for 3D gaze tracking tailored for mixed-reality settings, aimed at enhancing joint attention and collaborative efforts in team-based scenarios. Conventional gaze tracking, often limited by monocular cameras and traditional eye-tracking apparatus, struggles with simultaneous data synchronization and analysis from multiple participants in group contexts. Our proposed framework leverages state-of-the-art computer vision and machine learning techniques to overcome these obstacles, enabling precise 3D gaze estimation without dependence on specialized hardware or complex data fusion. Utilizing facial recognition and deep learning, the framework achieves real-time, tracking of gaze patterns across several individuals, addressing common depth estimation errors, and ensuring spatial and identity consistency within the dataset. Empirical results demonstrate the accuracy and reliability of our method in group environments. This provides mechanisms for significant advances in behavior and interaction analysis in educational and professional training applications in dynamic and unstructured environments., Comment: 9 pages, 8 figures, conference, submitted to ICMI 2024

We present Diffusion Soup, a compartmentalization method for Text-to-Image Generation that averages the weights of diffusion models trained on sharded data. By construction, our approach enables training-free continual learning and unlearning with no additional memory or inference costs, since models corresponding to data shards can be added or removed by re-averaging. We show that Diffusion Soup samples from a point in weight space that approximates the geometric mean of the distributions of constituent datasets, which offers anti-memorization guarantees and enables zero-shot style mixing. Empirically, Diffusion Soup outperforms a paragon model trained on the union of all data shards and achieves a 30% improvement in Image Reward (.34 $\to$ .44) on domain sharded data, and a 59% improvement in IR (.37 $\to$ .59) on aesthetic data. In both cases, souping also prevails in TIFA score (respectively, 85.5 $\to$ 86.5 and 85.6 $\to$ 86.8). We demonstrate robust unlearning -- removing any individual domain shard only lowers performance by 1% in IR (.45 $\to$ .44) -- and validate our theoretical insights on anti-memorization using real data. Finally, we showcase Diffusion Soup's ability to blend the distinct styles of models finetuned on different shards, resulting in the zero-shot generation of hybrid styles.

Over 85 million computed tomography (CT) scans are performed annually in the US, of which approximately one quarter focus on the abdomen. Given the current radiologist shortage, there is a large impetus to use artificial intelligence to alleviate the burden of interpreting these complex imaging studies. Prior state-of-the-art approaches for automated medical image interpretation leverage vision language models (VLMs). However, current medical VLMs are generally limited to 2D images and short reports, and do not leverage electronic health record (EHR) data for supervision. We introduce Merlin - a 3D VLM that we train using paired CT scans (6+ million images from 15,331 CTs), EHR diagnosis codes (1.8+ million codes), and radiology reports (6+ million tokens). We evaluate Merlin on 6 task types and 752 individual tasks. The non-adapted (off-the-shelf) tasks include zero-shot findings classification (31 findings), phenotype classification (692 phenotypes), and zero-shot cross-modal retrieval (image to findings and image to impressions), while model adapted tasks include 5-year disease prediction (6 diseases), radiology report generation, and 3D semantic segmentation (20 organs). We perform internal validation on a test set of 5,137 CTs, and external validation on 7,000 clinical CTs and on two public CT datasets (VerSe, TotalSegmentator). Beyond these clinically-relevant evaluations, we assess the efficacy of various network architectures and training strategies to depict that Merlin has favorable performance to existing task-specific baselines. We derive data scaling laws to empirically assess training data needs for requisite downstream task performance. Furthermore, unlike conventional VLMs that require hundreds of GPUs for training, we perform all training on a single GPU., Comment: 18 pages, 7 figures

Data driven models for automated diagnosis in radiology suffer from insufficient and imbalanced datasets due to low representation of pathology in a population and the cost of expert annotations. Datasets can be bolstered through data augmentation. However, even when utilizing a full suite of transformations during model training, typical data augmentations do not address variations in human anatomy. An alternative direction is to synthesize data using generative models, which can potentially craft datasets with specific attributes. While this holds promise, commonly used generative models such as Generative Adversarial Networks may inadvertently produce anatomically inaccurate features. On the other hand, diffusion models, which offer greater stability, tend to memorize training data, raising concerns about privacy and generative diversity. Alternatively, inpainting has the potential to augment data through directly inserting pathology in medical images. However, this approach introduces a new challenge: accurately merging the generated pathological features with the surrounding anatomical context. While inpainting is a well established method for addressing simple lesions, its application to pathologies that involve complex structural changes remains relatively unexplored. We propose an efficient method for inpainting pathological features onto healthy anatomy in MRI through voxelwise noise scheduling in a latent diffusion model. We evaluate the method's ability to insert disc herniation and central canal stenosis in lumbar spine sagittal T2 MRI, and it achieves superior Frechet Inception Distance compared to state-of-the-art methods.

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., Comment: In CVPR 2024

In this paper, we propose an effective two-stage approach named Grounded-Dreamer to generate 3D assets that can accurately follow complex, compositional text prompts while achieving high fidelity by using a pre-trained multi-view diffusion model. Multi-view diffusion models, such as MVDream, have shown to generate high-fidelity 3D assets using score distillation sampling (SDS). However, applied naively, these methods often fail to comprehend compositional text prompts, and may often entirely omit certain subjects or parts. To address this issue, we first advocate leveraging text-guided 4-view images as the bottleneck in the text-to-3D pipeline. We then introduce an attention refocusing mechanism to encourage text-aligned 4-view image generation, without the necessity to re-train the multi-view diffusion model or craft a high-quality compositional 3D dataset. We further propose a hybrid optimization strategy to encourage synergy between the SDS loss and the sparse RGB reference images. Our method consistently outperforms previous state-of-the-art (SOTA) methods in generating compositional 3D assets, excelling in both quality and accuracy, and enabling diverse 3D from the same text prompt., Comment: 9 pages, 10 figures

Existing unified image segmentation models either employ a unified architecture across multiple tasks but use separate weights tailored to each dataset, or apply a single set of weights to multiple datasets but are limited to a single task. In this paper, we introduce the Mixed-Query Transformer (MQ-Former), a unified architecture for multi-task and multi-dataset image segmentation using a single set of weights. To enable this, we propose a mixed query strategy, which can effectively and dynamically accommodate different types of objects without heuristic designs. In addition, the unified architecture allows us to use data augmentation with synthetic masks and captions to further improve model generalization. Experiments demonstrate that MQ-Former can not only effectively handle multiple segmentation datasets and tasks compared to specialized state-of-the-art models with competitive performance, but also generalize better to open-set segmentation tasks, evidenced by over 7 points higher performance than the prior art on the open-vocabulary SeginW benchmark.

Scaling up model and data size has been quite successful for the evolution of LLMs. However, the scaling law for the diffusion based text-to-image (T2I) models is not fully explored. It is also unclear how to efficiently scale the model for better performance at reduced cost. The different training settings and expensive training cost make a fair model comparison extremely difficult. In this work, we empirically study the scaling properties of diffusion based T2I models by performing extensive and rigours ablations on scaling both denoising backbones and training set, including training scaled UNet and Transformer variants ranging from 0.4B to 4B parameters on datasets upto 600M images. For model scaling, we find the location and amount of cross attention distinguishes the performance of existing UNet designs. And increasing the transformer blocks is more parameter-efficient for improving text-image alignment than increasing channel numbers. We then identify an efficient UNet variant, which is 45% smaller and 28% faster than SDXL's UNet. On the data scaling side, we show the quality and diversity of the training set matters more than simply dataset size. Increasing caption density and diversity improves text-image alignment performance and the learning efficiency. Finally, we provide scaling functions to predict the text-image alignment performance as functions of the scale of model size, compute and dataset size., Comment: CVPR2024

We delve into Open Domain Generalization (ODG), marked by domain and category shifts between training's labeled source and testing's unlabeled target domains. Existing solutions to ODG face limitations due to constrained generalizations of traditional CNN backbones and errors in detecting target open samples in the absence of prior knowledge. Addressing these pitfalls, we introduce ODG-CLIP, harnessing the semantic prowess of the vision-language model, CLIP. Our framework brings forth three primary innovations: Firstly, distinct from prevailing paradigms, we conceptualize ODG as a multi-class classification challenge encompassing both known and novel categories. Central to our approach is modeling a unique prompt tailored for detecting unknown class samples, and to train this, we employ a readily accessible stable diffusion model, elegantly generating proxy images for the open class. Secondly, aiming for domain-tailored classification (prompt) weights while ensuring a balance of precision and simplicity, we devise a novel visual stylecentric prompt learning mechanism. Finally, we infuse images with class-discriminative knowledge derived from the prompt space to augment the fidelity of CLIP's visual embeddings. We introduce a novel objective to safeguard the continuity of this infused semantic intel across domains, especially for the shared classes. Through rigorous testing on diverse datasets, covering closed and open-set DG contexts, ODG-CLIP demonstrates clear supremacy, consistently outpacing peers with performance boosts between 8%-16%. Code will be available at https://github.com/mainaksingha01/ODG-CLIP., Comment: Accepted in CVPR 2024

We present a generic Reinforcement Learning (RL) framework optimized for crafting adversarial attacks on different model types spanning from ECG signal analysis (1D), image classification (2D), and video classification (3D). The framework focuses on identifying sensitive regions and inducing misclassifications with minimal distortions and various distortion types. The novel RL method outperforms state-of-the-art methods for all three applications, proving its efficiency. Our RL approach produces superior localization masks, enhancing interpretability for image classification and ECG analysis models. For applications such as ECG analysis, our platform highlights critical ECG segments for clinicians while ensuring resilience against prevalent distortions. This comprehensive tool aims to bolster both resilience with adversarial training and transparency across varied applications and data types., Comment: AAAI Proceedings reference: https://ojs.aaai.org/index.php/AAAI/article/view/30579

Retrieval Augmented Generation (RAG) is emerging as a flexible and robust technique to adapt models to private users data without training, to handle credit attribution, and to allow efficient machine unlearning at scale. However, RAG techniques for image generation may lead to parts of the retrieved samples being copied in the model's output. To reduce risks of leaking private information contained in the retrieved set, we introduce Copy-Protected generation with Retrieval (CPR), a new method for RAG with strong copyright protection guarantees in a mixed-private setting for diffusion models.CPR allows to condition the output of diffusion models on a set of retrieved images, while also guaranteeing that unique identifiable information about those example is not exposed in the generated outputs. In particular, it does so by sampling from a mixture of public (safe) distribution and private (user) distribution by merging their diffusion scores at inference. We prove that CPR satisfies Near Access Freeness (NAF) which bounds the amount of information an attacker may be able to extract from the generated images. We provide two algorithms for copyright protection, CPR-KL and CPR-Choose. Unlike previously proposed rejection-sampling-based NAF methods, our methods enable efficient copyright-protected sampling with a single run of backward diffusion. We show that our method can be applied to any pre-trained conditional diffusion model, such as Stable Diffusion or unCLIP. In particular, we empirically show that applying CPR on top of unCLIP improves quality and text-to-image alignment of the generated results (81.4 to 83.17 on TIFA benchmark), while enabling credit attribution, copy-right protection, and deterministic, constant time, unlearning., Comment: CVPR 2024

Generative Vision-Language Models (VLMs) are prone to generate plausible-sounding textual answers that, however, are not always grounded in the input image. We investigate this phenomenon, usually referred to as "hallucination" and show that it stems from an excessive reliance on the language prior. In particular, we show that as more tokens are generated, the reliance on the visual prompt decreases, and this behavior strongly correlates with the emergence of hallucinations. To reduce hallucinations, we introduce Multi-Modal Mutual-Information Decoding (M3ID), a new sampling method for prompt amplification. M3ID amplifies the influence of the reference image over the language prior, hence favoring the generation of tokens with higher mutual information with the visual prompt. M3ID can be applied to any pre-trained autoregressive VLM at inference time without necessitating further training and with minimal computational overhead. If training is an option, we show that M3ID can be paired with Direct Preference Optimization (DPO) to improve the model's reliance on the prompt image without requiring any labels. Our empirical findings show that our algorithms maintain the fluency and linguistic capabilities of pre-trained VLMs while reducing hallucinations by mitigating visually ungrounded answers. Specifically, for the LLaVA 13B model, M3ID and M3ID+DPO reduce the percentage of hallucinated objects in captioning tasks by 25% and 28%, respectively, and improve the accuracy on VQA benchmarks such as POPE by 21% and 24%.

Neural Radiance Field (NeRF), as an implicit 3D scene representation, lacks inherent ability to accommodate changes made to the initial static scene. If objects are reconfigured, it is difficult to update the NeRF to reflect the new state of the scene without time-consuming data re-capturing and NeRF re-training. To address this limitation, we develop the first update method for NeRFs to physical changes. Our method takes only sparse new images (e.g. 4) of the altered scene as extra inputs and update the pre-trained NeRF in around 1 to 2 minutes. Particularly, we develop a pipeline to identify scene changes and update the NeRF accordingly. Our core idea is the use of a second helper NeRF to learn the local geometry and appearance changes, which sidesteps the optimization difficulties in direct NeRF fine-tuning. The interpolation power of the helper NeRF is the key to accurately reconstruct the un-occluded objects regions under sparse view supervision. Our method imposes no constraints on NeRF pre-training, and requires no extra user input or explicit semantic priors. It is an order of magnitude faster than re-training NeRF from scratch while maintaining on-par and even superior performance.

Industrial anomaly detection is an important task within computer vision with a wide range of practical use cases. The small size of anomalous regions in many real-world datasets necessitates processing the images at a high resolution. This frequently poses significant challenges concerning memory consumption during the model training and inference stages, leaving some existing methods impractical for widespread adoption. To overcome this challenge, we present the tiled ensemble approach, which reduces memory consumption by dividing the input images into a grid of tiles and training a dedicated model for each tile location. The tiled ensemble is compatible with any existing anomaly detection model without the need for any modification of the underlying architecture. By introducing overlapping tiles, we utilize the benefits of traditional stacking ensembles, leading to further improvements in anomaly detection capabilities beyond high resolution alone. We perform a comprehensive analysis using diverse underlying architectures, including Padim, PatchCore, FastFlow, and Reverse Distillation, on two standard anomaly detection datasets: MVTec and VisA. Our method demonstrates a notable improvement across setups while remaining within GPU memory constraints, consuming only as much GPU memory as a single model needs to process a single tile., Comment: To appear at CVPR 24 Visual Anomaly Detection Workshop. Research conducted during Google Summer of Code 2023 at OpenVINO (Intel). GSoC 2023 page: https://summerofcode.withgoogle.com/archive/2023/projects/WUSjdxGl

This report introduces a new family of multimodal models, Gemini, that exhibit remarkable capabilities across image, audio, video, and text understanding. The Gemini family consists of Ultra, Pro, and Nano sizes, suitable for applications ranging from complex reasoning tasks to on-device memory-constrained use-cases. Evaluation on a broad range of benchmarks shows that our most-capable Gemini Ultra model advances the state of the art in 30 of 32 of these benchmarks - notably being the first model to achieve human-expert performance on the well-studied exam benchmark MMLU, and improving the state of the art in every one of the 20 multimodal benchmarks we examined. We believe that the new capabilities of the Gemini family in cross-modal reasoning and language understanding will enable a wide variety of use cases. We discuss our approach toward post-training and deploying Gemini models responsibly to users through services including Gemini, Gemini Advanced, Google AI Studio, and Cloud Vertex AI.

The development of generative models that create 3D content from a text prompt has made considerable strides thanks to the use of the score distillation sampling (SDS) method on pre-trained diffusion models for image generation. However, the SDS method is also the source of several artifacts, such as the Janus problem, the misalignment between the text prompt and the generated 3D model, and 3D model inaccuracies. While existing methods heavily rely on the qualitative assessment of these artifacts through visual inspection of a limited set of samples, in this work we propose more objective quantitative evaluation metrics, which we cross-validate via human ratings, and show analysis of the failure cases of the SDS technique. We demonstrate the effectiveness of this analysis by designing a novel computationally efficient baseline model that achieves state-of-the-art performance on the proposed metrics while addressing all the above-mentioned artifacts.

Map representation learned by expert demonstrations has shown promising research value. However, recent advancements in the visual navigation field face challenges due to the lack of human datasets in the real world for efficient supervised representation learning of the environments. We present a Landmark-Aware Visual Navigation (LAVN) dataset to allow for supervised learning of human-centric exploration policies and map building. We collect RGB observation and human point-click pairs as a human annotator explores virtual and real-world environments with the goal of full coverage exploration of the space. The human annotators also provide distinct landmark examples along each trajectory, which we intuit will simplify the task of map or graph building and localization. These human point-clicks serve as direct supervision for waypoint prediction when learning to explore in environments. Our dataset covers a wide spectrum of scenes, including rooms in indoor environments, as well as walkways outdoors. Dataset is available at DOI: 10.5281/zenodo.10608067.

Visual navigation follows the intuition that humans can navigate without detailed maps. A common approach is interactive exploration while building a topological graph with images at nodes that can be used for planning. Recent variations learn from passive videos and can navigate using complex social and semantic cues. However, a significant number of training videos are needed, large graphs are utilized, and scenes are not unseen since odometry is utilized. We introduce a new approach to visual navigation using feudal learning, which employs a hierarchical structure consisting of a worker agent, a mid-level manager, and a high-level manager. Key to the feudal learning paradigm, agents at each level see a different aspect of the task and operate at different spatial and temporal scales. Two unique modules are developed in this framework. For the high-level manager, we learn a memory proxy map in a self supervised manner to record prior observations in a learned latent space and avoid the use of graphs and odometry. For the mid-level manager, we develop a waypoint network that outputs intermediate subgoals imitating human waypoint selection during local navigation. This waypoint network is pre-trained using a new, small set of teleoperation videos that we make publicly available, with training environments different from testing environments. The resulting feudal navigation network achieves near SOTA performance, while providing a novel no-RL, no-graph, no-odometry, no-metric map approach to the image goal navigation task.

Visually-conditioned language models (VLMs) have seen growing adoption in applications such as visual dialogue, scene understanding, and robotic task planning; adoption that has fueled a wealth of new models such as LLaVa, InstructBLIP, and PaLI-3. Despite the volume of new releases, key design decisions around image preprocessing, architecture, and optimization are under-explored, making it challenging to understand what factors account for model performance $-$ a challenge further complicated by the lack of objective, consistent evaluations. To address these gaps, we first compile a suite of standardized evaluations spanning visual question answering, object localization, and challenge sets that probe properties such as hallucination; evaluations that provide fine-grained insight VLM capabilities. Second, we rigorously investigate VLMs along key design axes, including pretrained visual representations and training from base vs. instruct-tuned language models, amongst others. We couple our analysis with three resource contributions: (1) a unified framework for evaluating VLMs, (2) optimized, flexible training code, and (3) checkpoints for all models, including a family of VLMs at the 7-13B scale that strictly outperform InstructBLIP and LLaVa v1.5, the state-of-the-art in open VLMs., Comment: Published at ICML 2024. 22 pages, 11 figures. Training code and models: https://github.com/TRI-ML/prismatic-vlms. Evaluation code: https://github.com/TRI-ML/vlm-evaluation

Recent advances in visual anomaly detection research have seen AUROC and AUPRO scores on public benchmark datasets such as MVTec and VisA converge towards perfect recall, giving the impression that these benchmarks are near-solved. However, high AUROC and AUPRO scores do not always reflect qualitative performance, which limits the validity of these metrics in real-world applications. We argue that the artificial ceiling imposed by the lack of an adequate evaluation metric restrains progression of the field, and it is crucial that we revisit the evaluation metrics used to rate our algorithms. In response, we introduce Per-IMage Overlap (PIMO), a novel metric that addresses the shortcomings of AUROC and AUPRO. PIMO retains the recall-based nature of the existing metrics but introduces two distinctions: the assignment of curves (and respective area under the curve) is per-image, and its X-axis relies solely on normal images. Measuring recall per image simplifies instance score indexing and is more robust to noisy annotations. As we show, it also accelerates computation and enables the usage of statistical tests to compare models. By imposing low tolerance for false positives on normal images, PIMO provides an enhanced model validation procedure and highlights performance variations across datasets. Our experiments demonstrate that PIMO offers practical advantages and nuanced performance insights that redefine anomaly detection benchmarks -- notably challenging the perception that MVTec AD and VisA datasets have been solved by contemporary models. Available on GitHub: https://github.com/jpcbertoldo/aupimo., Comment: Accepted to BMVC 2024. Official implementation: https://github.com/jpcbertoldo/aupimo. Integrated in anomalib https://github.com/openvinotoolkit/anomalib. This research was conducted during Google Summer of Code 2023 (GSoC 2023) with the anomalib team from Intel's OpenVINO Toolkit

Sleep detection and annotation are crucial for researchers to understand sleep patterns, especially in children. With modern wrist-worn watches comprising built-in accelerometers, sleep logs can be collected. However, the annotation of these logs into distinct sleep events: onset and wakeup, proves to be challenging. These annotations must be automated, precise, and scalable. We propose to model the accelerometer data using different machine learning (ML) techniques such as support vectors, boosting, ensemble methods, and more complex approaches involving LSTMs and Region-based CNNs. Later, we aim to evaluate these approaches using the Event Detection Average Precision (EDAP) score (similar to the IOU metric) to eventually compare the predictive power and model performance.

We present a novel framework for generating adversarial benchmarks to evaluate the robustness of image classification models. Our framework allows users to customize the types of distortions to be optimally applied to images, which helps address the specific distortions relevant to their deployment. The benchmark can generate datasets at various distortion levels to assess the robustness of different image classifiers. Our results show that the adversarial samples generated by our framework with any of the image classification models, like ResNet-50, Inception-V3, and VGG-16, are effective and transferable to other models causing them to fail. These failures happen even when these models are adversarially retrained using state-of-the-art techniques, demonstrating the generalizability of our adversarial samples. We achieve competitive performance in terms of net $L_2$ distortion compared to state-of-the-art benchmark techniques on CIFAR-10 and ImageNet; however, we demonstrate our framework achieves such results with simple distortions like Gaussian noise without introducing unnatural artifacts or color bleeds. This is made possible by a model-based reinforcement learning (RL) agent and a technique that reduces a deep tree search of the image for model sensitivity to perturbations, to a one-level analysis and action. The flexibility of choosing distortions and setting classification probability thresholds for multiple classes makes our framework suitable for algorithmic audits., Comment: 2023 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)

Tracking subjects in videos is one of the most widely used functions in camera-based IoT applications such as security surveillance, smart city traffic safety enhancement, vehicle to pedestrian communication and so on. In the computer vision domain, tracking is usually achieved by first detecting subjects with bounding boxes, then associating detected bounding boxes across video frames. For many IoT systems, images captured by cameras are usually sent over the network to be processed at a different site that has more powerful computing resources than edge devices. However, sending entire frames through the network causes significant bandwidth consumption that may exceed the system bandwidth constraints. To tackle this problem, we propose ViFiT, a transformer-based model that reconstructs vision bounding box trajectories from phone data (IMU and Fine Time Measurements). It leverages a transformer ability of better modeling long-term time series data. ViFiT is evaluated on Vi-Fi Dataset, a large-scale multimodal dataset in 5 diverse real world scenes, including indoor and outdoor environments. To fill the gap of proper metrics of jointly capturing the system characteristics of both tracking quality and video bandwidth reduction, we propose a novel evaluation framework dubbed Minimum Required Frames (MRF) and Minimum Required Frames Ratio (MRFR). ViFiT achieves an MRFR of 0.65 that outperforms the state-of-the-art approach for cross-modal reconstruction in LSTM Encoder-Decoder architecture X-Translator of 0.98, resulting in a high frame reduction rate as 97.76%., Comment: 22 pages, 12 figures, 9 tables. MobiCom 2023 ISACom

The generation of effective latent representations and their subsequent refinement to incorporate precise information is an essential prerequisite for Vision-Language Understanding (VLU) tasks such as Video Question Answering (VQA). However, most existing methods for VLU focus on sparsely sampling or fine-graining the input information (e.g., sampling a sparse set of frames or text tokens), or adding external knowledge. We present a novel "DRAX: Distraction Removal and Attended Cross-Alignment" method to rid our cross-modal representations of distractors in the latent space. We do not exclusively confine the perception of any input information from various modalities but instead use an attention-guided distraction removal method to increase focus on task-relevant information in latent embeddings. DRAX also ensures semantic alignment of embeddings during cross-modal fusions. We evaluate our approach on a challenging benchmark (SUTD-TrafficQA dataset), testing the framework's abilities for feature and event queries, temporal relation understanding, forecasting, hypothesis, and causal analysis through extensive experiments.

Practitioners interested in using computer vision models lack user-friendly and open-source software that combines features to label training data, allow multiple users, train new algorithms, review output, and implement new models. Labeling training data, such as images, is a key step to developing accurate object detection algorithms using computer vision. This step is often not compatible with many cloud-based services for marking or labeling image and video data due to limited internet bandwidth in many regions of the world. Desktop tools are useful for groups working in remote locations, but users often do not have the capability to combine projects developed locally by multiple collaborators. Furthermore, many tools offer features for labeling data or using pre-trained models for classification, but few allow researchers to combine these steps to create and apply custom models. Free, open-source, and user-friendly software that offers a full suite of features (e.g., ability to work locally and online, and train custom models) is desirable to field researchers and conservationists that may have limited coding skills. We developed Njobvu-AI, a free, open-source tool that can be run on both desktop and server hardware using Node.js, allowing users to label data, combine projects for collaboration and review, train custom algorithms, and implement new computer vision models. The name Njobvu-AI (pronounced N-joh-voo AI), incorporating the Chichewa word for elephant, is inspired by a wildlife monitoring program in Malawi that was a primary impetus for the development of this tool and references similarities between the powerful memory of elephants and properties of computer vision models., Comment: 13 pages, 6 figures. For code and documentation, see https://github.com/sullichrosu/Njobvu-AI/

We introduce Compartmentalized Diffusion Models (CDM), a method to train different diffusion models (or prompts) on distinct data sources and arbitrarily compose them at inference time. The individual models can be trained in isolation, at different times, and on different distributions and domains and can be later composed to achieve performance comparable to a paragon model trained on all data simultaneously. Furthermore, each model only contains information about the subset of the data it was exposed to during training, enabling several forms of training data protection. In particular, CDMs enable perfect selective forgetting and continual learning for large-scale diffusion models, allow serving customized models based on the user's access rights. Empirically the quality (FID) of the class-conditional CDMs (8-splits) is within 10% (on fine-grained vision datasets) of a monolithic model (no splits), and allows (8x) faster forgetting compared monolithic model with a maximum FID increase of 1%. When applied to text-to-image generation, CDMs improve alignment (TIFA) by 14.33% over a monolithic model trained on MSCOCO. CDMs also allow determining the importance of a subset of the data (attribution) in generating particular samples, and reduce memorization.

This paper presents Volume-DROID, a novel approach for Simultaneous Localization and Mapping (SLAM) that integrates Volumetric Mapping and Differentiable Recurrent Optimization-Inspired Design (DROID). Volume-DROID takes camera images (monocular or stereo) or frames from a video as input and combines DROID-SLAM, point cloud registration, an off-the-shelf semantic segmentation network, and Convolutional Bayesian Kernel Inference (ConvBKI) to generate a 3D semantic map of the environment and provide accurate localization for the robot. The key innovation of our method is the real-time fusion of DROID-SLAM and Convolutional Bayesian Kernel Inference (ConvBKI), achieved through the introduction of point cloud generation from RGB-Depth frames and optimized camera poses. This integration, engineered to enable efficient and timely processing, minimizes lag and ensures effective performance of the system. Our approach facilitates functional real-time online semantic mapping with just camera images or stereo video input. Our paper offers an open-source Python implementation of the algorithm, available at https://github.com/peterstratton/Volume-DROID.

We describe a first step towards learning general-purpose visual representations of physical scenes using only image prediction as a training criterion. To do so, we first define "physical scene" and show that, even though different agents may maintain different representations of the same scene, the underlying physical scene that can be inferred is unique. Then, we show that NeRFs cannot represent the physical scene, as they lack extrapolation mechanisms. Those, however, could be provided by Diffusion Models, at least in theory. To test this hypothesis empirically, NeRFs can be combined with Diffusion Models, a process we refer to as NeRF Diffusion, used as unsupervised representations of the physical scene. Our analysis is limited to visual data, without external grounding mechanisms that can be provided by independent sensory modalities., Comment: TLDR: Physical scenes are equivalence classes of sufficient statistics, and can be inferred uniquely by any agent measuring the same finite data; We formalize and implement an approach to representation learning that overturns "naive realism" in favor of an analytical approach of Russell and Koenderink. NeRFs cannot capture the physical scenes, but combined with Diffusion Models they can

This research focuses on real-time surveillance systems as a means for tackling the issue of non-compliance with helmet regulations, a practice that considerably amplifies the risk for motorcycle drivers or riders. Despite the well-established advantages of helmet usage, achieving widespread compliance remains challenging due to diverse contributing factors. To effectively address this concern, real-time monitoring and enforcement of helmet laws have been proposed as a plausible solution. However, previous attempts at real-time helmet violation detection have been hindered by their limited ability to operate in real-time. To overcome this limitation, the current paper introduces a novel real-time helmet violation detection system that utilizes the YOLOv5 single-stage object detection model. This model is trained on the 2023 NVIDIA AI City Challenge 2023 Track 5 dataset. The optimal hyperparameters for training the model are determined using genetic algorithms. Additionally, data augmentation and various sampling techniques are implemented to enhance the model's performance. The efficacy of the models is evaluated using precision, recall, and mean Average Precision (mAP) metrics. The results demonstrate impressive precision, recall, and mAP scores of 0.848, 0.599, and 0.641, respectively for the training data. Furthermore, the model achieves notable mAP score of 0.6667 for the test datasets, leading to a commendable 4th place rank in the public leaderboard. This innovative approach represents a notable breakthrough in the field and holds immense potential to substantially enhance motorcycle safety. By enabling real-time monitoring and enforcement capabilities, this system has the capacity to contribute towards increased compliance with helmet laws, thereby effectively reducing the risks faced by motorcycle riders and passengers.

We present a novel self-contained camera-projector tabletop system with a lamp form-factor that brings digital intelligence to our tables. We propose a real-time, on-device, learning-based touch detection algorithm that makes any tabletop interactive. The top-down configuration and learning-based algorithm makes our method robust to the presence of clutter, a main limitation of existing camera-projector tabletop systems. Our research prototype enables a set of experiences that combine hand interactions and objects present on the table. A video can be found at https://youtu.be/hElC_c25Fg8.

Prompt learning is an efficient approach to adapt transformers by inserting learnable set of parameters into the input and intermediate representations of a pre-trained model. In this work, we present Expressive Prompts with Residuals (EXPRES) which modifies the prompt learning paradigm specifically for effective adaptation of vision transformers (ViT). Out method constructs downstream representations via learnable ``output'' tokens, that are akin to the learned class tokens of the ViT. Further for better steering of the downstream representation processed by the frozen transformer, we introduce residual learnable tokens that are added to the output of various computations. We apply EXPRES for image classification, few shot learning, and semantic segmentation, and show our method is capable of achieving state of the art prompt tuning on 3/3 categories of the VTAB benchmark. In addition to strong performance, we observe that our approach is an order of magnitude more prompt efficient than existing visual prompting baselines. We analytically show the computational benefits of our approach over weight space adaptation techniques like finetuning. Lastly we systematically corroborate the architectural design of our method via a series of ablation experiments., Comment: Accepted at CVPR (2023)

Biometrics plays a significant role in vision-based surveillance applications. Soft biometrics such as gait is widely used with face in surveillance tasks like person recognition and re-identification. Nevertheless, in practical scenarios, classical fusion techniques respond poorly to changes in individual users and in the external environment. To this end, we propose a novel adaptive multi-biometric fusion strategy for the dynamic incorporation of gait and face biometric cues by leveraging keyless attention deep neural networks. Various external factors such as viewpoint and distance to the camera, are investigated in this study. Extensive experiments have shown superior performanceof the proposed model compared with the state-of-the-art model.

We propose InCA, a lightweight method for transfer learning that cross-attends to any activation layer of a pre-trained model. During training, InCA uses a single forward pass to extract multiple activations, which are passed to external cross-attention adapters, trained anew and combined or selected for downstream tasks. We show that, even when selecting a single top-scoring adapter, InCA achieves performance comparable to full fine-tuning, at a cost comparable to fine-tuning just the last layer. For example, with a cross-attention probe 1.3% the size of a pre-trained ViT-L/16 model, we achieve performance within 0.2% of the full fine-tuning paragon at a computational training cost of 51% of the baseline, on average across 11 downstream classification. Unlike other forms of efficient adaptation, InCA does not require backpropagating through the pre-trained model, thus leaving its execution unaltered at both training and inference. The versatility of InCA is best illustrated in fine-grained tasks, which may require accessing information absent in the last layer but accessible in intermediate layer activations. Since the backbone is fixed, InCA allows parallel ensembling as well as parallel execution of multiple tasks. InCA achieves state-of-the-art performance in the ImageNet-to-Sketch multi-task benchmark., Comment: Accepted to NeurIPS 2023

Deep neural network (DNN) models for retinopathy have estimated predictive accuracies in the mid-to-high 90%. However, the following aspects remain unaddressed: State-of-the-art models are complex and require substantial computational infrastructure to train and deploy; The reliability of predictions can vary widely. In this paper, we focus on these aspects and propose a form of iterative knowledge distillation(IKD), called IKD+ that incorporates a tradeoff between size, accuracy and reliability. We investigate the functioning of IKD+ using two widely used techniques for estimating model calibration (Platt-scaling and temperature-scaling), using the best-performing model available, which is an ensemble of EfficientNets with approximately 100M parameters. We demonstrate that IKD+ equipped with temperature-scaling results in models that show up to approximately 500-fold decreases in the number of parameters than the original ensemble without a significant loss in accuracy. In addition, calibration scores (reliability) for the IKD+ models are as good as or better than the base mode, Comment: Submitted to IEEE International Conference on Image Processing (ICIP 2023)

We propose an approach to estimate the number of samples required for a model to reach a target performance. We find that the power law, the de facto principle to estimate model performance, leads to large error when using a small dataset (e.g., 5 samples per class) for extrapolation. This is because the log-performance error against the log-dataset size follows a nonlinear progression in the few-shot regime followed by a linear progression in the high-shot regime. We introduce a novel piecewise power law (PPL) that handles the two data regimes differently. To estimate the parameters of the PPL, we introduce a random forest regressor trained via meta learning that generalizes across classification/detection tasks, ResNet/ViT based architectures, and random/pre-trained initializations. The PPL improves the performance estimation on average by 37% across 16 classification and 33% across 10 detection datasets, compared to the power law. We further extend the PPL to provide a confidence bound and use it to limit the prediction horizon that reduces over-estimation of data by 76% on classification and 91% on detection datasets., Comment: CVPR 2023

Whole Slide Images (WSIs) or histopathology images are used in digital pathology. WSIs pose great challenges to deep learning models for clinical diagnosis, owing to their size and lack of pixel-level annotations. With the recent advancements in computational pathology, newer multiple-instance learning-based models have been proposed. Multiple-instance learning for WSIs necessitates creating patches and uses the encoding of these patches for diagnosis. These models use generic pre-trained models (ResNet-50 pre-trained on ImageNet) for patch encoding. The recently proposed KimiaNet, a DenseNet121 model pre-trained on TCGA slides, is a domain-specific pre-trained model. This paper shows the effect of domain-specific pre-training on WSI classification. To investigate the effect of domain-specific pre-training, we considered the current state-of-the-art multiple-instance learning models, 1) CLAM, an attention-based model, and 2) TransMIL, a self-attention-based model, and evaluated the models' confidence and predictive performance in detecting primary brain tumors - gliomas. Domain-specific pre-training improves the confidence of the models and also achieves a new state-of-the-art performance of WSI-based glioma subtype classification, showing a high clinical applicability in assisting glioma diagnosis. We will publicly share our code and experimental results at https://github.com/soham-chitnis10/WSI-domain-specific., Comment: Accepted in EMBC 2023

Herein, we present Stratus, an end-to-end full-stack deep learning application deployed on the cloud. The rise of productionized deep learning necessitates infrastructure in the cloud that can provide such service (IaaS). In this paper, we explore the use of modern cloud infrastructure and micro-services to deliver accurate and high-speed predictions to an end-user, using a Deep Neural Network (DNN) to predict handwritten digit input, interfaced via a full-stack application. We survey tooling from Spark ML, Apache Kafka, Chameleon Cloud, Ansible, Vagrant, Python Flask, Docker, and Kubernetes in order to realize this machine learning pipeline. Through our cloud-based approach, we are able to demonstrate benchmark performance on the MNIST dataset with a deep learning model.

In this paper, we propose and showcase, for the first time, monocular multi-view layout estimation for warehouse racks and shelves. Unlike typical layout estimation methods, MVRackLay estimates multi-layered layouts, wherein each layer corresponds to the layout of a shelf within a rack. Given a sequence of images of a warehouse scene, a dual-headed Convolutional-LSTM architecture outputs segmented racks, the front and the top view layout of each shelf within a rack. With minimal effort, such an output is transformed into a 3D rendering of all racks, shelves and objects on the shelves, giving an accurate 3D depiction of the entire warehouse scene in terms of racks, shelves and the number of objects on each shelf. MVRackLay generalizes to a diverse set of warehouse scenes with varying number of objects on each shelf, number of shelves and in the presence of other such racks in the background. Further, MVRackLay shows superior performance vis-a-vis its single view counterpart, RackLay, in layout accuracy, quantized in terms of the mean IoU and mAP metrics. We also showcase a multi-view stitching of the 3D layouts resulting in a representation of the warehouse scene with respect to a global reference frame akin to a rendering of the scene from a SLAM pipeline. To the best of our knowledge, this is the first such work to portray a 3D rendering of a warehouse scene in terms of its semantic components - Racks, Shelves and Objects - all from a single monocular camera.

The current approach to marking attendance in colleges is tedious and time consuming. I propose AttenFace, a standalone system to analyze, track and grant attendance in real time using face recognition. Using snapshots of class from live camera feed, the system identifies students and marks them as present in a class based on their presence in multiple snapshots taken throughout the class duration. Face recognition for each class is performed independently and in parallel, ensuring that the system scales with number of concurrent classes. Further, the separation of the face recognition server from the back-end server for attendance calculation allows the face recognition module to be integrated with existing attendance tracking software like Moodle. The face recognition algorithm runs at 10 minute intervals on classroom snapshots, significantly reducing computation compared to direct processing of live camera feed. This method also provides students the flexibility to leave class for a short duration (such as for a phone call) without losing attendance for that class. Attendance is granted to a student if he remains in class for a number of snapshots above a certain threshold. The system is fully automatic and requires no professor intervention or any form of manual attendance or even camera set-up, since the back-end directly interfaces with in-class cameras. AttenFace is a first-of-its-kind one-stop solution for face-recognition-enabled attendance in educational institutions that prevents proxy, handling all aspects from students checking attendance to professors deciding their own attendance policy, to college administration enforcing default attendance rules., Comment: 5 pages, 8 figures. To be published in conference proceedings of IEEE Conference on Information and Communication Technology (CICT) 2022

Today's architectural engineering and construction (AEC) software require a learning curve to generate a three-dimension building representation. This limits the ability to quickly validate the volumetric implications of an initial design idea communicated via a single sketch. Allowing designers to translate a single sketch to a 3D building will enable owners to instantly visualize 3D project information without the cognitive load required. If previous state-of-the-art (SOTA) data-driven methods for single view reconstruction (SVR) showed outstanding results in the reconstruction process from a single image or sketch, they lacked specific applications, analysis, and experiments in the AEC. Therefore, this research addresses this gap, introducing the first deep learning method focused only on buildings that aim to convert a single sketch to a 3D building mesh: Vitruvio. Vitruvio adapts Occupancy Network for SVR tasks on a specific building dataset (Manhattan 1K). This adaptation brings two main improvements. First, it accelerates the inference process by more than 26% (from 0.5s to 0.37s). Second, it increases the reconstruction accuracy (measured by the Chamfer Distance) by 18%. During this adaptation in the AEC domain, we evaluate the effect of the building orientation in the learning procedure since it constitutes an important design factor. While aligning all the buildings to a canonical pose improved the overall quantitative metrics, it did not capture fine-grain details in more complex building shapes (as shown in our qualitative analysis). Finally, Vitruvio outputs a 3D-printable building mesh with arbitrary topology and genus from a single perspective sketch, providing a step forward to allow owners and designers to communicate 3D information via a 2D, effective, intuitive, and universal communication medium: the sketch.

We introduce ViFiCon, a self-supervised contrastive learning scheme which uses synchronized information across vision and wireless modalities to perform cross-modal association. Specifically, the system uses pedestrian data collected from RGB-D camera footage as well as WiFi Fine Time Measurements (FTM) from a user's smartphone device. We represent the temporal sequence by stacking multi-person depth data spatially within a banded image. Depth data from RGB-D (vision domain) is inherently linked with an observable pedestrian, but FTM data (wireless domain) is associated only to a smartphone on the network. To formulate the cross-modal association problem as self-supervised, the network learns a scene-wide synchronization of the two modalities as a pretext task, and then uses that learned representation for the downstream task of associating individual bounding boxes to specific smartphones, i.e. associating vision and wireless information. We use a pre-trained region proposal model on the camera footage and then feed the extrapolated bounding box information into a dual-branch convolutional neural network along with the FTM data. We show that compared to fully supervised SoTA models, ViFiCon achieves high performance vision-to-wireless association, finding which bounding box corresponds to which smartphone device, without hand-labeled association examples for training data.

Social media apps have become very promising and omnipresent in daily life. Most social media apps are used to deliver vital information to those nearby and far away. As our lives become more hectic, many of us strive to limit our usage of social media apps because they are too addictive, and the majority of us have gotten preoccupied with our daily lives. Because of this, we frequently overlook crucial information, such as invitations to weddings, interviews, birthday parties, etc., or find ourselves unable to attend the event. In most cases, this happens because users are more likely to discover the invitation or information only before the event, giving them little time to prepare. To solve this issue, in this study, we created a system that will collect social media chat and filter it using Natural Language Processing (NLP) methods like Tokenization, Stop Words Removal, Lemmatization, Segmentation, and Named Entity Recognition (NER). Also, Machine Learning Algorithms such as K-Nearest Neighbor (KNN) Algorithm are implemented to prioritize the received invitation and to sort the level of priority. Finally, a customized notification will be delivered to the users where they acknowledge the upcoming event. So, the chances of missing the event are less or can be planned., Comment: 4 pages, 5 figures

Mathematical reasoning, a core ability of human intelligence, presents unique challenges for machines in abstract thinking and logical reasoning. Recent large pre-trained language models such as GPT-3 have achieved remarkable progress on mathematical reasoning tasks written in text form, such as math word problems (MWP). However, it is unknown if the models can handle more complex problems that involve math reasoning over heterogeneous information, such as tabular data. To fill the gap, we present Tabular Math Word Problems (TabMWP), a new dataset containing 38,431 open-domain grade-level problems that require mathematical reasoning on both textual and tabular data. Each question in TabMWP is aligned with a tabular context, which is presented as an image, semi-structured text, and a structured table. There are two types of questions: free-text and multi-choice, and each problem is annotated with gold solutions to reveal the multi-step reasoning process. We evaluate different pre-trained models on TabMWP, including the GPT-3 model in a few-shot setting. As earlier studies suggest, since few-shot GPT-3 relies on the selection of in-context examples, its performance is unstable and can degrade to near chance. The unstable issue is more severe when handling complex problems like TabMWP. To mitigate this, we further propose a novel approach, PromptPG, which utilizes policy gradient to learn to select in-context examples from a small amount of training data and then constructs the corresponding prompt for the test example. Experimental results show that our method outperforms the best baseline by 5.31% on the accuracy metric and reduces the prediction variance significantly compared to random selection, which verifies its effectiveness in selecting in-context examples., Comment: ICLR 2023. 26 pages and 18 figures. The data and code are available at https://promptpg.github.io