We investigate the CLIP image encoder by analyzing how individual model components affect the final representation. We decompose the image representation as a sum across individual image patches, model layers, and attention heads, and use CLIP's text representation to interpret the summands. Interpreting the attention heads, we characterize each head's role by automatically finding text representations that span its output space, which reveals property-specific roles for many heads (e.g. location or shape). Next, interpreting the image patches, we uncover an emergent spatial localization within CLIP. Finally, we use this understanding to remove spurious features from CLIP and to create a strong zero-shot image segmenter. Our results indicate that a scalable understanding of transformer models is attainable and can be used to repair and improve models.
相關內容
Reconstructing visual stimuli from functional Magnetic Resonance Imaging (fMRI) based on Latent Diffusion Models (LDM) provides a fine-grained retrieval of the brain. A challenge persists in reconstructing a cohesive alignment of details (such as structure, background, texture, color, etc.). Moreover, LDMs would generate different image results even under the same conditions. For these, we first uncover the neuroscientific perspective of LDM-based methods that is top-down creation based on pre-trained knowledge from massive images but lack of detail-driven bottom-up perception resulting in unfaithful details. We propose NeuralDiffuser which introduces primary visual feature guidance to provide detail cues in the form of gradients, extending the bottom-up process for LDM-based methods to achieve faithful semantics and details. We also developed a novel guidance strategy to ensure the consistency of repeated reconstructions rather than a variety of results. We obtain the state-of-the-art performance of NeuralDiffuser on the Natural Senses Dataset (NSD), which offers more faithful details and consistent results.
Many recent advancements in recommender systems have focused on developing sequence-based and graph-based approaches. Both approaches proved useful in modeling intricate relationships within behavioral data, leading to promising outcomes in personalized ranking and next-item recommendation tasks while maintaining good scalability. However, they capture very different signals from data. While the former approach represents users directly through ordered interactions with recent items, the latter one aims to capture indirect dependencies across the interactions graph. This paper presents a novel multi-representational learning framework that exploits the synergies between these two paradigms. Our empirical evaluation on several datasets demonstrates that mutual training of sequential and graph components with the proposed framework significantly improves recommendations performance.
Monocular 3D detection (M3D) aims for precise 3D object localization from a single-view image which usually involves labor-intensive annotation of 3D detection boxes. Weakly supervised M3D has recently been studied to obviate the 3D annotation process by leveraging many existing 2D annotations, but it often requires extra training data such as LiDAR point clouds or multi-view images which greatly degrades its applicability and usability in various applications. We propose SKD-WM3D, a weakly supervised monocular 3D detection framework that exploits depth information to achieve M3D with a single-view image exclusively without any 3D annotations or other training data. One key design in SKD-WM3D is a self-knowledge distillation framework, which transforms image features into 3D-like representations by fusing depth information and effectively mitigates the inherent depth ambiguity in monocular scenarios with little computational overhead in inference. In addition, we design an uncertainty-aware distillation loss and a gradient-targeted transfer modulation strategy which facilitate knowledge acquisition and knowledge transfer, respectively. Extensive experiments show that SKD-WM3D surpasses the state-of-the-art clearly and is even on par with many fully supervised methods.
This work presents a new method for enhancing communication efficiency in stochastic Federated Learning that trains over-parameterized random networks. In this setting, a binary mask is optimized instead of the model weights, which are kept fixed. The mask characterizes a sparse sub-network that is able to generalize as good as a smaller target network. Importantly, sparse binary masks are exchanged rather than the floating point weights in traditional federated learning, reducing communication cost to at most 1 bit per parameter (Bpp). We show that previous state of the art stochastic methods fail to find sparse networks that can reduce the communication and storage overhead using consistent loss objectives. To address this, we propose adding a regularization term to local objectives that acts as a proxy of the transmitted masks entropy, therefore encouraging sparser solutions by eliminating redundant features across sub-networks. Extensive empirical experiments demonstrate significant improvements in communication and memory efficiency of up to five magnitudes compared to the literature, with minimal performance degradation in validation accuracy in some instances
Large Language Models (LLMs) are a class of generative AI models built using the Transformer network, capable of leveraging vast datasets to identify, summarize, translate, predict, and generate language. LLMs promise to revolutionize society, yet training these foundational models poses immense challenges. Semantic vector search within large language models is a potent technique that can significantly enhance search result accuracy and relevance. Unlike traditional keyword-based search methods, semantic search utilizes the meaning and context of words to grasp the intent behind queries and deliver more precise outcomes. Elasticsearch emerges as one of the most popular tools for implementing semantic search an exceptionally scalable and robust search engine designed for indexing and searching extensive datasets. In this article, we delve into the fundamentals of semantic search and explore how to harness Elasticsearch and Transformer models to bolster large language model processing paradigms. We gain a comprehensive understanding of semantic search principles and acquire practical skills for implementing semantic search in real-world model application scenarios.
With the extremely rapid advances in remote sensing (RS) technology, a great quantity of Earth observation (EO) data featuring considerable and complicated heterogeneity is readily available nowadays, which renders researchers an opportunity to tackle current geoscience applications in a fresh way. With the joint utilization of EO data, much research on multimodal RS data fusion has made tremendous progress in recent years, yet these developed traditional algorithms inevitably meet the performance bottleneck due to the lack of the ability to comprehensively analyse and interpret these strongly heterogeneous data. Hence, this non-negligible limitation further arouses an intense demand for an alternative tool with powerful processing competence. Deep learning (DL), as a cutting-edge technology, has witnessed remarkable breakthroughs in numerous computer vision tasks owing to its impressive ability in data representation and reconstruction. Naturally, it has been successfully applied to the field of multimodal RS data fusion, yielding great improvement compared with traditional methods. This survey aims to present a systematic overview in DL-based multimodal RS data fusion. More specifically, some essential knowledge about this topic is first given. Subsequently, a literature survey is conducted to analyse the trends of this field. Some prevalent sub-fields in the multimodal RS data fusion are then reviewed in terms of the to-be-fused data modalities, i.e., spatiospectral, spatiotemporal, light detection and ranging-optical, synthetic aperture radar-optical, and RS-Geospatial Big Data fusion. Furthermore, We collect and summarize some valuable resources for the sake of the development in multimodal RS data fusion. Finally, the remaining challenges and potential future directions are highlighted.
We present a large-scale study on unsupervised spatiotemporal representation learning from videos. With a unified perspective on four recent image-based frameworks, we study a simple objective that can easily generalize all these methods to space-time. Our objective encourages temporally-persistent features in the same video, and in spite of its simplicity, it works surprisingly well across: (i) different unsupervised frameworks, (ii) pre-training datasets, (iii) downstream datasets, and (iv) backbone architectures. We draw a series of intriguing observations from this study, e.g., we discover that encouraging long-spanned persistency can be effective even if the timespan is 60 seconds. In addition to state-of-the-art results in multiple benchmarks, we report a few promising cases in which unsupervised pre-training can outperform its supervised counterpart. Code is made available at //github.com/facebookresearch/SlowFast
Video captioning is a challenging task that requires a deep understanding of visual scenes. State-of-the-art methods generate captions using either scene-level or object-level information but without explicitly modeling object interactions. Thus, they often fail to make visually grounded predictions, and are sensitive to spurious correlations. In this paper, we propose a novel spatio-temporal graph model for video captioning that exploits object interactions in space and time. Our model builds interpretable links and is able to provide explicit visual grounding. To avoid unstable performance caused by the variable number of objects, we further propose an object-aware knowledge distillation mechanism, in which local object information is used to regularize global scene features. We demonstrate the efficacy of our approach through extensive experiments on two benchmarks, showing our approach yields competitive performance with interpretable predictions.
Recent advancements in deep neural networks for graph-structured data have led to state-of-the-art performance on recommender system benchmarks. However, making these methods practical and scalable to web-scale recommendation tasks with billions of items and hundreds of millions of users remains a challenge. Here we describe a large-scale deep recommendation engine that we developed and deployed at Pinterest. We develop a data-efficient Graph Convolutional Network (GCN) algorithm PinSage, which combines efficient random walks and graph convolutions to generate embeddings of nodes (i.e., items) that incorporate both graph structure as well as node feature information. Compared to prior GCN approaches, we develop a novel method based on highly efficient random walks to structure the convolutions and design a novel training strategy that relies on harder-and-harder training examples to improve robustness and convergence of the model. We also develop an efficient MapReduce model inference algorithm to generate embeddings using a trained model. We deploy PinSage at Pinterest and train it on 7.5 billion examples on a graph with 3 billion nodes representing pins and boards, and 18 billion edges. According to offline metrics, user studies and A/B tests, PinSage generates higher-quality recommendations than comparable deep learning and graph-based alternatives. To our knowledge, this is the largest application of deep graph embeddings to date and paves the way for a new generation of web-scale recommender systems based on graph convolutional architectures.
Medical image segmentation requires consensus ground truth segmentations to be derived from multiple expert annotations. A novel approach is proposed that obtains consensus segmentations from experts using graph cuts (GC) and semi supervised learning (SSL). Popular approaches use iterative Expectation Maximization (EM) to estimate the final annotation and quantify annotator's performance. Such techniques pose the risk of getting trapped in local minima. We propose a self consistency (SC) score to quantify annotator consistency using low level image features. SSL is used to predict missing annotations by considering global features and local image consistency. The SC score also serves as the penalty cost in a second order Markov random field (MRF) cost function optimized using graph cuts to derive the final consensus label. Graph cut obtains a global maximum without an iterative procedure. Experimental results on synthetic images, real data of Crohn's disease patients and retinal images show our final segmentation to be accurate and more consistent than competing methods.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191