Few-shot object detection (FSOD), an efficient method for addressing the severe data-hungry problem, has been extensively discussed. Current works have significantly advanced the problem in terms of model and data. However, the overall performance of most FSOD methods still does not fulfill the desired accuracy. In this paper we improve the FSOD model to address the severe issue of sample imbalance and weak feature propagation. To alleviate modeling bias from data-sufficient base classes, we examine the effect of decoupling the parameters for classes with sufficient data and classes with few samples in various ways. We design a base-novel categories decoupled DETR (DeDETR) for FSOD. We also explore various types of skip connection between the encoder and decoder for DETR. Besides, we notice that the best outputs could come from the intermediate layer of the decoder instead of the last layer; therefore, we build a unified decoder module that could dynamically fuse the decoder layers as the output feature. We evaluate our model on commonly used datasets such as PASCAL VOC and MSCOCO. Our results indicate that our proposed module could achieve stable improvements of 5% to 10% in both fine-tuning and meta-learning paradigms and has outperformed the highest score in recent works.
Change detection is a critical task in earth observation applications. Recently, deep learning-based methods have shown promising performance and are quickly adopted in change detection. However, the widely used multiple encoder and single decoder (MESD) as well as dual encoder-decoder (DED) architectures still struggle to effectively handle change detection well. The former has problems of bitemporal feature interference in the feature-level fusion, while the latter is inapplicable to intraclass change detection and multiview building change detection. To solve these problems, we propose a new strategy with an exchanging dual encoder-decoder structure for binary change detection with semantic guidance and spatial localization. The proposed strategy solves the problems of bitemporal feature inference in MESD by fusing bitemporal features in the decision level and the inapplicability in DED by determining changed areas using bitemporal semantic features. We build a binary change detection model based on this strategy, and then validate and compare it with 18 state-of-the-art change detection methods on six datasets in three scenarios, including intraclass change detection datasets (CDD, SYSU), single-view building change detection datasets (WHU, LEVIR-CD, LEVIR-CD+) and a multiview building change detection dataset (NJDS). The experimental results demonstrate that our model achieves superior performance with high efficiency and outperforms all benchmark methods with F1-scores of 97.77%, 83.07%, 94.86%, 92.33%, 91.39%, 74.35% on CDD, SYSU, WHU, LEVIR-CD, LEVIR- CD+, and NJDS datasets, respectively. The code of this work will be available at https://github.com/NJU-LHRS/official-SGSLN.
Speaker adaptation systems face privacy concerns, for such systems are trained on private datasets and often overfitting. This paper demonstrates that an attacker can extract speaker information by querying speaker-adapted speech recognition (ASR) systems. We focus on the speaker information of a transformer-based ASR and propose GhostVec, a simple and efficient attack method to extract the speaker information from an encoder-decoder-based ASR system without any external speaker verification system or natural human voice as a reference. To make our results quantitative, we pre-process GhostVec using singular value decomposition (SVD) and synthesize it into waveform. Experiment results show that the synthesized audio of GhostVec reaches 10.83\% EER and 0.47 minDCF with target speakers, which suggests the effectiveness of the proposed method. We hope the preliminary discovery in this study to catalyze future speech recognition research on privacy-preserving topics.
Transformer-based pretrained large language models (PLM) such as BERT and GPT have achieved remarkable success in NLP tasks. However, PLMs are prone to encoding stereotypical biases. Although a burgeoning literature has emerged on stereotypical bias mitigation in PLMs, such as work on debiasing gender and racial stereotyping, how such biases manifest and behave internally within PLMs remains largely unknown. Understanding the internal stereotyping mechanisms may allow better assessment of model fairness and guide the development of effective mitigation strategies. In this work, we focus on attention heads, a major component of the Transformer architecture, and propose a bias analysis framework to explore and identify a small set of biased heads that are found to contribute to a PLM's stereotypical bias. We conduct extensive experiments to validate the existence of these biased heads and to better understand how they behave. We investigate gender and racial bias in the English language in two types of Transformer-based PLMs: the encoder-based BERT model and the decoder-based autoregressive GPT model. Overall, the results shed light on understanding the bias behavior in pretrained language models.
Phenotype-based screening has attracted much attention for identifying cell-active compounds. Transcriptional and proteomic profiles of cell population or single cells are informative phenotypic measures of cellular responses to perturbations. In this paper, we proposed a deep learning framework based on encoder-decoder architecture that maps the initial cellular states to a latent space, in which we assume the effects of drug perturbation on cellular states follow linear additivity. Next, we introduced the cycle consistency constraints to enforce that initial cellular state subjected to drug perturbations would produce the perturbed cellular responses, and, conversely, removal of drug perturbation from the perturbed cellular states would restore the initial cellular states. The cycle consistency constraints and linear modeling in latent space enable to learn interpretable and transferable drug perturbation representations, so that our model can predict cellular response to unseen drugs. We validated our model on three different types of datasets, including bulk transcriptional responses, bulk proteomic responses, and single-cell transcriptional responses to drug perturbations. The experimental results show that our model achieves better performance than existing state-of-the-art methods.
Supervised learning algorithms based on Convolutional Neural Networks have become the benchmark for medical image segmentation tasks, but their effectiveness heavily relies on a large amount of labeled data. However, annotating medical image datasets is a laborious and time-consuming process. Inspired by semi-supervised algorithms that use both labeled and unlabeled data for training, we propose the PLGDF framework, which builds upon the mean teacher network for segmenting medical images with less annotation. We propose a novel pseudo-label utilization scheme, which combines labeled and unlabeled data to augment the dataset effectively. Additionally, we enforce the consistency between different scales in the decoder module of the segmentation network and propose a loss function suitable for evaluating the consistency. Moreover, we incorporate a sharpening operation on the predicted results, further enhancing the accuracy of the segmentation. Extensive experiments on three publicly available datasets demonstrate that the PLGDF framework can largely improve performance by incorporating the unlabeled data. Meanwhile, our framework yields superior performance compared to six state-of-the-art semi-supervised learning methods. The codes of this study are available at https://github.com/ortonwang/PLGDF.
Text and vision foundation models can perform many tasks in a zero-shot setting, a desirable property that enables these systems to be applied in general and low-resource settings. There has been far less work, however, on the zero-shot abilities of ASR foundation models, with these systems typically fine-tuned to specific tasks or constrained to applications that match their training criterion and data annotation. In this work we investigate the ability of Whisper and MMS, ASR foundation models trained primarily for speech recognition, to perform zero-shot audio classification. We use simple template-based text prompts at the decoder and use the resulting decoding probabilities to generate zero-shot predictions. Without training the model on extra data or adding any new parameters, we demonstrate that Whisper shows promising zero-shot classification performance on a range of 8 audio-classification datasets, outperforming the accuracy of existing state-of-the-art zero-shot baselines by an average of 9%. One important step to unlock the emergent ability is debiasing, where a simple unsupervised reweighting method of the class probabilities yields consistent significant performance gains. We further show that performance increases with model size, implying that as ASR foundation models scale up, they may exhibit improved zero-shot performance.
We propose a theoretical framework for formulating language model decoder algorithms with dynamic programming and information theory. With dynamic programming, we lift the design of decoder algorithms from the logit space to the action-state value function space, and show that the decoding algorithms are consequences of optimizing the action-state value functions. Each component in the action-state value function space has an information theoretical interpretation. With the lifting and interpretation, it becomes evident what the decoder algorithm is optimized for, and hence facilitating the arbitration of the tradeoffs in sensibleness, diversity, and attribution.
Generating human language through non-invasive brain-computer interfaces (BCIs) has the potential to unlock many applications, such as serving disabled patients and improving communication. Currently, however, generating language via BCIs has been previously successful only within a classification setup for selecting pre-generated sentence continuation candidates with the most likely cortical semantic representation. Inspired by recent research that revealed associations between the brain and the large computational language models, we propose a generative language BCI that utilizes the capacity of a large language model (LLM) jointly with a semantic brain decoder to directly generate language from functional magnetic resonance imaging (fMRI) input. The proposed model can generate coherent language sequences aligned with the semantic content of visual or auditory language stimuli perceived, without prior knowledge of any pre-generated candidates. We compare the language generated from the presented model with a random control, pre-generated language selection approach, and a standard LLM, which generates common coherent text solely based on the next word likelihood according to statistical language training data. The proposed model is found to generate language that is more aligned with semantic stimulus in response to which brain input is sampled. Our findings demonstrate the potential and feasibility of employing BCIs in direct language generation.
Due to the limited availability of high quality datasets for training sentence embeddings in Turkish, we propose a training methodology and a regimen to develop a sentence embedding model. The central idea is simple but effective : is to fine-tune a pretrained encoder-decoder model in two consecutive stages, where the first stage involves aligning the embedding space with translation pairs. Thanks to this alignment, the prowess of the main model can be better projected onto the target language in a sentence embedding setting where it can be fine-tuned with high accuracy in short duration with limited target language dataset.
Despite the prevalence of tabular datasets, few-shot learning remains under-explored within this domain. Existing few-shot methods are not directly applicable to tabular datasets due to varying column relationships, meanings, and permutational invariance. To address these challenges, we propose FLAT-a novel approach to tabular few-shot learning, encompassing knowledge sharing between datasets with heterogeneous feature spaces. Utilizing an encoder inspired by Dataset2Vec, FLAT learns low-dimensional embeddings of datasets and their individual columns, which facilitate knowledge transfer and generalization to previously unseen datasets. A decoder network parametrizes the predictive target network, implemented as a Graph Attention Network, to accommodate the heterogeneous nature of tabular datasets. Experiments on a diverse collection of 118 UCI datasets demonstrate FLAT's successful generalization to new tabular datasets and a considerable improvement over the baselines.
The interaction decoder utilized in prevalent Transformer-based HOI detectors typically accepts pre-composed human-object pairs as inputs. Though achieving remarkable performance, such paradigm lacks feasibility and cannot explore novel combinations over entities during decoding. We present L OGIC HOI, a new HOI detector that leverages neural-logic reasoning and Transformer to infer feasible interactions between entities. Specifically, we modify the self-attention mechanism in vanilla Transformer, enabling it to reason over the <human, action, object> triplet and constitute novel interactions. Meanwhile, such reasoning process is guided by two crucial properties for understanding HOI: affordances (the potential actions an object can facilitate) and proxemics (the spatial relations between humans and objects). We formulate these two properties in first-order logic and ground them into continuous space to constrain the learning process of our approach, leading to improved performance and zero-shot generalization capabilities. We evaluate L OGIC HOI on V-COCO and HICO-DET under both normal and zero-shot setups, achieving significant improvements over existing methods.
Inverse rendering is an ill-posed problem. Previous work has sought to resolve this by focussing on priors for object or scene shape or appearance. In this work, we instead focus on a prior for natural illuminations. Current methods rely on spherical harmonic lighting or other generic representations and, at best, a simplistic prior on the parameters. This results in limitations for the inverse setting in terms of the expressivity of the illumination conditions, especially when taking specular reflections into account. We propose a conditional neural field representation based on a variational auto-decoder and a transformer decoder. We extend Vector Neurons to build equivariance directly into our architecture, and leveraging insights from depth estimation through a scale-invariant loss function, we enable the accurate representation of High Dynamic Range (HDR) images. The result is a compact, rotation-equivariant HDR neural illumination model capable of capturing complex, high-frequency features in natural environment maps. Training our model on a curated dataset of 1.6K HDR environment maps of natural scenes, we compare it against traditional representations, demonstrate its applicability for an inverse rendering task and show environment map completion from partial observations. We share our PyTorch implementation, dataset and trained models at https://github.com/JADGardner/ns_reni
Many large language models (LLMs) for medicine have largely been evaluated on short texts, and their ability to handle longer sequences such as a complete electronic health record (EHR) has not been systematically explored. Assessing these models on long sequences is crucial since prior work in the general domain has demonstrated performance degradation of LLMs on longer texts. Motivated by this, we introduce LongBoX, a collection of seven medical datasets in text-to-text format, designed to investigate model performance on long sequences. Preliminary experiments reveal that both medical LLMs (e.g., BioGPT) and strong general domain LLMs (e.g., FLAN-T5) struggle on this benchmark. We further evaluate two techniques designed for long-sequence handling: (i) local-global attention, and (ii) Fusion-in-Decoder (FiD). Our results demonstrate mixed results with long-sequence handling - while scores on some datasets increase, there is substantial room for improvement. We hope that LongBoX facilitates the development of more effective long-sequence techniques for the medical domain. Data and source code are available at https://github.com/Mihir3009/LongBoX.
Blind detection of the forged regions in digital images is an effective authentication means to counter the malicious use of local image editing techniques. Existing encoder-decoder forensic networks overlook the fact that detecting complex and subtle tampered regions typically requires more feedback information. In this paper, we propose a Progressive FeedbACk-enhanced Transformer (ProFact) network to achieve coarse-to-fine image forgery localization. Specifically, the coarse localization map generated by an initial branch network is adaptively fed back to the early transformer encoder layers, which can enhance the representation of positive features while suppressing interference factors. The cascaded transformer network, combined with a contextual spatial pyramid module, is designed to refine discriminative forensic features for improving the forgery localization accuracy and reliability. Furthermore, we present an effective strategy to automatically generate large-scale forged image samples close to real-world forensic scenarios, especially in realistic and coherent processing. Leveraging on such samples, a progressive and cost-effective two-stage training protocol is applied to the ProFact network. The extensive experimental results on nine public forensic datasets show that our proposed localizer greatly outperforms the state-of-the-art on the generalization ability and robustness of image forgery localization. Code will be publicly available at https://github.com/multimediaFor/ProFact.
Although image captioning has a vast array of applications, it has not reached its full potential in languages other than English. Arabic, for instance, although the native language of more than 400 million people, remains largely underrepresented in this area. This is due to the lack of labeled data and powerful Arabic generative models. We alleviate this issue by presenting a novel vision-language model dedicated to Arabic, dubbed \textit{Violet}. Our model is based on a vision encoder and a Gemini text decoder that maintains generation fluency while allowing fusion between the vision and language components. To train our model, we introduce a new method for automatically acquiring data from available English datasets. We also manually prepare a new dataset for evaluation. \textit{Violet} performs sizeably better than our baselines on all of our evaluation datasets. For example, it reaches a CIDEr score of $61.2$ on our manually annotated dataset and achieves an improvement of $13$ points on Flickr8k.
Video post-processing methods can improve the quality of compressed videos at the decoder side. Most of the existing methods need to train corresponding models for compressed videos with different quantization parameters to improve the quality of compressed videos. However, in most cases, the quantization parameters of the decoded video are unknown. This makes existing methods have their limitations in improving video quality. To tackle this problem, this work proposes a diffusion model based post-processing method for compressed videos. The proposed method first estimates the feature vectors of the compressed video and then uses the estimated feature vectors as the prior information for the quality enhancement model to adaptively enhance the quality of compressed video with different quantization parameters. Experimental results show that the quality enhancement results of our proposed method on mixed datasets are superior to existing methods.
Multilingual pre-trained Large Language Models (LLMs) are incredibly effective at Question Answering (QA), a core task in Natural Language Understanding, achieving high accuracies on several multilingual benchmarks. However, little is known about how well their confidences are calibrated. In this paper, we comprehensively benchmark the calibration of several multilingual LLMs (MLLMs) on a variety of QA tasks. We perform extensive experiments, spanning encoder-only, encoder-decoder, and decoder-only QA models (size varying from 110M to 7B parameters) and diverse languages, including both high- and low-resource ones. We study different dimensions of calibration in in-distribution, out-of-distribution, and cross-lingual transfer settings, and investigate strategies to improve it, including post-hoc methods and regularized fine-tuning. For decoder-only LLMs such as LlaMa2, we additionally find that in-context learning improves confidence calibration on multilingual data. We also conduct several ablation experiments to study the effect of language distances, language corpus size, and model size on calibration, and how multilingual models compare with their monolingual counterparts for diverse tasks and languages. Our experiments suggest that the multilingual QA models are poorly calibrated for languages other than English and incorporating a small set of cheaply translated multilingual samples during fine-tuning/calibration effectively enhances the calibration performance.
Encoder-decoder transformer models have achieved great success on various vision-language (VL) tasks, but they suffer from high inference latency. Typically, the decoder takes up most of the latency because of the auto-regressive decoding. To accelerate the inference, we propose an approach of performing Dynamic Early Exit on Decoder (DEED). We build a multi-exit encoder-decoder transformer model which is trained with deep supervision so that each of its decoder layers is capable of generating plausible predictions. In addition, we leverage simple yet practical techniques, including shared generation head and adaptation modules, to keep accuracy when exiting at shallow decoder layers. Based on the multi-exit model, we perform step-level dynamic early exit during inference, where the model may decide to use fewer decoder layers based on its confidence of the current layer at each individual decoding step. Considering different number of decoder layers may be used at different decoding steps, we compute deeper-layer decoder features of previous decoding steps just-in-time, which ensures the features from different decoding steps are semantically aligned. We evaluate our approach with two state-of-the-art encoder-decoder transformer models on various VL tasks. We show our approach can reduce overall inference latency by 30%-60% with comparable or even higher accuracy compared to baselines.
We present Multiple-Question Multiple-Answer (MQMA), a novel approach to do text-VQA in encoder-decoder transformer models. The text-VQA task requires a model to answer a question by understanding multi-modal content: text (typically from OCR) and an associated image. To the best of our knowledge, almost all previous approaches for text-VQA process a single question and its associated content to predict a single answer. In order to answer multiple questions from the same image, each question and content are fed into the model multiple times. In contrast, our proposed MQMA approach takes multiple questions and content as input at the encoder and predicts multiple answers at the decoder in an auto-regressive manner at the same time. We make several novel architectural modifications to standard encoder-decoder transformers to support MQMA. We also propose a novel MQMA denoising pre-training task which is designed to teach the model to align and delineate multiple questions and content with associated answers. MQMA pre-trained model achieves state-of-the-art results on multiple text-VQA datasets, each with strong baselines. Specifically, on OCR-VQA (+2.5%), TextVQA (+1.4%), ST-VQA (+0.6%), DocVQA (+1.1%) absolute improvements over the previous state-of-the-art approaches.