We show how to "compile" human-readable programs into standard decoder-only transformer models. Our compiler, Tracr, generates models with known structure. This structure can be used to design experiments. For example, we use it to study "superposition" in transformers that execute multi-step algorithms. Additionally, the known structure of Tracr-compiled models can serve as ground-truth for evaluating interpretability methods. Commonly, because the "programs" learned by transformers are unknown it is unclear whether an interpretation succeeded. We demonstrate our approach by implementing and examining programs including computing token frequencies, sorting, and parenthesis checking. We provide an open-source implementation of Tracr at https://github.com/google-deepmind/tracr.
The emergence of industrial-scale speech recognition (ASR) models such as Whisper and USM, trained on 1M hours of weakly labelled and 12M hours of audio only proprietary data respectively, has led to a stronger need for large scale public ASR corpora and competitive open source pipelines. Unlike the said models, large language models are typically based on Transformer decoders, and it remains unclear if decoder-only models trained on public data alone can deliver competitive performance. In this work, we investigate factors such as choice of training datasets and modeling components necessary for obtaining the best performance using public English ASR corpora alone. Our Decoder-Only Transformer for ASR (DOTA) model comprehensively outperforms the encoder-decoder open source replication of Whisper (OWSM) on nearly all English ASR benchmarks and outperforms Whisper large-v3 on 7 out of 15 test sets. We release our codebase and model checkpoints under permissive license.
In recent years, the foundation models have swept the computer vision field and facilitated the development of various tasks within different modalities. However, it remains an open question on how to design an infrared foundation model. In this paper, we propose InfMAE, a foundation model in infrared modality. We release an infrared dataset, called Inf30 to address the problem of lacking large-scale data for self-supervised learning in the infrared vision community. Besides, we design an information-aware masking strategy, which is suitable for infrared images. This masking strategy allows for a greater emphasis on the regions with richer information in infrared images during the self-supervised learning process, which is conducive to learning the generalized representation. In addition, we adopt a multi-scale encoder to enhance the performance of the pre-trained encoders in downstream tasks. Finally, based on the fact that infrared images do not have a lot of details and texture information, we design an infrared decoder module, which further improves the performance of downstream tasks. Extensive experiments show that our proposed method InfMAE outperforms other supervised methods and self-supervised learning methods in three downstream tasks.
Feature attribution methods (FAs), such as gradients and attention, are widely employed approaches to derive the importance of all input features to the model predictions. Existing work in natural language processing has mostly focused on developing and testing FAs for encoder-only language models (LMs) in classification tasks. However, it is unknown if it is faithful to use these FAs for decoder-only models on text generation, due to the inherent differences between model architectures and task settings respectively. Moreover, previous work has demonstrated that there is no `one-wins-all' FA across models and tasks. This makes the selection of a FA computationally expensive for large LMs since input importance derivation often requires multiple forward and backward passes including gradient computations that might be prohibitive even with access to large compute. To address these issues, we present a model-agnostic FA for generative LMs called Recursive Attribution Generator (ReAGent). Our method updates the token importance distribution in a recursive manner. For each update, we compute the difference in the probability distribution over the vocabulary for predicting the next token between using the original input and using a modified version where a part of the input is replaced with RoBERTa predictions. Our intuition is that replacing an important token in the context should have resulted in a larger change in the model's confidence in predicting the token than replacing an unimportant token. Our method can be universally applied to any generative LM without accessing internal model weights or additional training and fine-tuning, as most other FAs require. We extensively compare the faithfulness of ReAGent with seven popular FAs across six decoder-only LMs of various sizes. The results show that our method consistently provides more faithful token importance distributions.
The 2024 ICASSP Auditory EEG Signal Processing Grand Challenge concerns the decoding of electroencephalography (EEG) measurements taken from participants who listened to speech material. This work details our solution to the match-mismatch sub-task: given a short temporal segment of EEG recordings and several candidate speech segments, the task is to classify which of the speech segments was time-aligned with the EEG signals. We show that high-frequency gamma-band responses to the speech envelope can be detected with a high accuracy. By jointly assessing gamma-band responses and low-frequency envelope tracking, we develop a match-mismatch decoder which placed first in this task.
State-space models are a low-complexity alternative to transformers for encoding long sequences and capturing long-term dependencies. We propose LOCOST: an encoder-decoder architecture based on state-space models for conditional text generation with long context inputs. With a computational complexity of $O(L \log L)$, this architecture can handle significantly longer sequences than state-of-the-art models that are based on sparse attention patterns. We evaluate our model on a series of long document abstractive summarization tasks. The model reaches a performance level that is 93-96% comparable to the top-performing sparse transformers of the same size while saving up to 50% memory during training and up to 87% during inference. Additionally, LOCOST effectively handles input texts exceeding 600K tokens at inference time, setting new state-of-the-art results on full-book summarization and opening new perspectives for long input processing.
Neural image compression has made a great deal of progress. State-of-the-art models are based on variational autoencoders and are outperforming classical models. Neural compression models learn to encode an image into a quantized latent representation that can be efficiently sent to the decoder, which decodes the quantized latent into a reconstructed image. While these models have proven successful in practice, they lead to sub-optimal results due to imperfect optimization and limitations in the encoder and decoder capacity. Recent work shows how to use stochastic Gumbel annealing (SGA) to refine the latents of pre-trained neural image compression models. We extend this idea by introducing SGA+, which contains three different methods that build upon SGA. We show how our method improves the overall compression performance in terms of the R-D trade-off, compared to its predecessors. Additionally, we show how refinement of the latents with our best-performing method improves the compression performance on both the Tecnick and CLIC dataset. Our method is deployed for a pre-trained hyperprior and for a more flexible model. Further, we give a detailed analysis of our proposed methods and show that they are less sensitive to hyperparameter choices. Finally, we show how each method can be extended to three- instead of two-class rounding.
Recent advancements in language models have significantly enhanced performance in multiple speech-related tasks. Existing speech language models typically utilize task-dependent prompt tokens to unify various speech tasks in a single model. However, this design omits the intrinsic connections between different speech tasks, which can potentially boost the performance of each task. In this work, we propose a novel decoder-only speech language model, SpeechComposer, that can unify common speech tasks by composing a fixed set of prompt tokens. Built upon four primary tasks -- speech synthesis, speech recognition, speech language modeling, and text language modeling -- SpeechComposer can easily extend to more speech tasks via compositions of well-designed prompt tokens, like voice conversion and speech enhancement. The unification of prompt tokens also makes it possible for knowledge sharing among different speech tasks in a more structured manner. Experimental results demonstrate that our proposed SpeechComposer can improve the performance of both primary tasks and composite tasks, showing the effectiveness of the shared prompt tokens. Remarkably, the unified decoder-only model achieves a comparable and even better performance than the baselines which are expert models designed for single tasks.
The rapidly changing landscape of technology and industries leads to dynamic skill requirements, making it crucial for employees and employers to anticipate such shifts to maintain a competitive edge in the labor market. Existing efforts in this area either rely on domain-expert knowledge or regarding skill evolution as a simplified time series forecasting problem. However, both approaches overlook the sophisticated relationships among different skills and the inner-connection between skill demand and supply variations. In this paper, we propose a Cross-view Hierarchical Graph learning Hypernetwork (CHGH) framework for joint skill demand-supply prediction. Specifically, CHGH is an encoder-decoder network consisting of i) a cross-view graph encoder to capture the interconnection between skill demand and supply, ii) a hierarchical graph encoder to model the co-evolution of skills from a cluster-wise perspective, and iii) a conditional hyper-decoder to jointly predict demand and supply variations by incorporating historical demand-supply gaps. Extensive experiments on three real-world datasets demonstrate the superiority of the proposed framework compared to seven baselines and the effectiveness of the three modules.
In this report, we explore the potential for text diffusion to replace autoregressive (AR) decoding for the training and deployment of large language models (LLMs). We are particularly interested to see whether pretrained AR models can be transformed into text diffusion models through a lightweight adaptation procedure we call ``AR2Diff''. We begin by establishing a strong baseline setup for training text diffusion models. Comparing across multiple architectures and pretraining objectives, we find that training a decoder-only model with a prefix LM objective is best or near-best across several tasks. Building on this finding, we test various transfer learning setups for text diffusion models. On machine translation, we find that text diffusion underperforms the standard AR approach. However, on code synthesis and extractive QA, we find diffusion models trained from scratch outperform AR models in many cases. We also observe quality gains from AR2Diff -- adapting AR models to use diffusion decoding. These results are promising given that text diffusion is relatively underexplored and can be significantly faster than AR decoding for long text generation.
Extracting meaningful information from high-dimensional data poses a formidable modeling challenge, particularly when the data is obscured by noise or represented through different modalities. This research proposes a novel non-parametric modeling approach, leveraging the Gaussian process (GP), to characterize high-dimensional data by mapping it to a latent low-dimensional manifold. This model, named the latent discriminative generative decoder (LDGD), employs both the data and associated labels in the manifold discovery process. We derive a Bayesian solution to infer the latent variables, allowing LDGD to effectively capture inherent stochasticity in the data. We demonstrate applications of LDGD on both synthetic and benchmark datasets. Not only does LDGD infer the manifold accurately, but its accuracy in predicting data points' labels surpasses state-of-the-art approaches. In the development of LDGD, we have incorporated inducing points to reduce the computational complexity of Gaussian processes for large datasets, enabling batch training for enhanced efficient processing and scalability. Additionally, we show that LDGD can robustly infer manifold and precisely predict labels for scenarios in that data size is limited, demonstrating its capability to efficiently characterize high-dimensional data with limited samples. These collective attributes highlight the importance of developing non-parametric modeling approaches to analyze high-dimensional data.
Leveraging rich information is crucial for dense prediction tasks. Light field (LF) cameras are instrumental in this regard, as they allow data to be sampled from various perspectives. This capability provides valuable spatial, depth, and angular information, enhancing scene-parsing tasks. However, we have identified two overlooked issues for the LF salient object detection (SOD) task. (1): Previous approaches predominantly employ a customized two-stream design to discover the spatial and depth features within light field images. The network struggles to learn the implicit angular information between different images due to a lack of intra-network data connectivity. (2): Little research has been directed towards the data augmentation strategy for LF SOD. Research on inter-network data connectivity is scant. In this study, we propose an efficient paradigm (LF Tracy) to address those issues. This comprises a single-pipeline encoder paired with a highly efficient information aggregation (IA) module (around 8M parameters) to establish an intra-network connection. Then, a simple yet effective data augmentation strategy called MixLD is designed to bridge the inter-network connections. Owing to this innovative paradigm, our model surpasses the existing state-of-the-art method through extensive experiments. Especially, LF Tracy demonstrates a 23% improvement over previous results on the latest large-scale PKU dataset. The source code is publicly available at: https://github.com/FeiBryantkit/LF-Tracy.
The remarkable prowess of diffusion models in image generation has spurred efforts to extend their application beyond generative tasks. However, a persistent challenge exists in lacking a unified approach to apply diffusion models to visual perception tasks with diverse semantic granularity requirements. Our purpose is to establish a unified visual perception framework, capitalizing on the potential synergies between generative and discriminative models. In this paper, we propose Vermouth, a simple yet effective framework comprising a pre-trained Stable Diffusion (SD) model containing rich generative priors, a unified head (U-head) capable of integrating hierarchical representations, and an adapted expert providing discriminative priors. Comprehensive investigations unveil potential characteristics of Vermouth, such as varying granularity of perception concealed in latent variables at distinct time steps and various U-net stages. We emphasize that there is no necessity for incorporating a heavyweight or intricate decoder to transform diffusion models into potent representation learners. Extensive comparative evaluations against tailored discriminative models showcase the efficacy of our approach on zero-shot sketch-based image retrieval (ZS-SBIR), few-shot classification, and open-vocabulary semantic segmentation tasks. The promising results demonstrate the potential of diffusion models as formidable learners, establishing their significance in furnishing informative and robust visual representations.
In this paper, we introduce YONOS-SR, a novel stable diffusion-based approach for image super-resolution that yields state-of-the-art results using only a single DDIM step. We propose a novel scale distillation approach to train our SR model. Instead of directly training our SR model on the scale factor of interest, we start by training a teacher model on a smaller magnification scale, thereby making the SR problem simpler for the teacher. We then train a student model for a higher magnification scale, using the predictions of the teacher as a target during the training. This process is repeated iteratively until we reach the target scale factor of the final model. The rationale behind our scale distillation is that the teacher aids the student diffusion model training by i) providing a target adapted to the current noise level rather than using the same target coming from ground truth data for all noise levels and ii) providing an accurate target as the teacher has a simpler task to solve. We empirically show that the distilled model significantly outperforms the model trained for high scales directly, specifically with few steps during inference. Having a strong diffusion model that requires only one step allows us to freeze the U-Net and fine-tune the decoder on top of it. We show that the combination of spatially distilled U-Net and fine-tuned decoder outperforms state-of-the-art methods requiring 200 steps with only one single step.
Quadrotors are increasingly used in the evolving field of aerial robotics for their agility and mechanical simplicity. However, inherent uncertainties, such as aerodynamic effects coupled with quadrotors' operation in dynamically changing environments, pose significant challenges for traditional, nominal model-based control designs. We propose a multi-task meta-learning method called Encoder-Prototype-Decoder (EPD), which has the advantage of effectively balancing shared and distinctive representations across diverse training tasks. Subsequently, we integrate the EPD model into a model predictive control problem (Proto-MPC) to enhance the quadrotor's ability to adapt and operate across a spectrum of dynamically changing tasks with an efficient online implementation. We validate the proposed method in simulations, which demonstrates Proto-MPC's robust performance in trajectory tracking of a quadrotor being subject to static and spatially varying side winds.
Accurate and automated gland segmentation on pathological images can assist pathologists in diagnosing the malignancy of colorectal adenocarcinoma. However, due to various gland shapes, severe deformation of malignant glands, and overlapping adhesions between glands. Gland segmentation has always been very challenging. To address these problems, we propose a DEA model. This model consists of two branches: the backbone encoding and decoding network and the local semantic extraction network. The backbone encoding and decoding network extracts advanced Semantic features, uses the proposed feature decoder to restore feature space information, and then enhances the boundary features of the gland through boundary enhancement attention. The local semantic extraction network uses the pre-trained DeepLabv3+ as a Local semantic-guided encoder to realize the extraction of edge features. Experimental results on two public datasets, GlaS and CRAG, confirm that the performance of our method is better than other gland segmentation methods.
Ultrasound measurements have been examined as potential tools for predicting the likelihood of successful vaginal delivery. The angle of progression (AoP) is a measurable parameter that can be obtained during the initial stage of labor. The AoP is defined as the angle between a straight line along the longitudinal axis of the pubic symphysis (PS) and a line from the inferior edge of the PS to the leading edge of the fetal head (FH). However, the process of measuring AoP on ultrasound images is time consuming and prone to errors. To address this challenge, we propose the Mix Transformer U-Net (MiTU-Net) network, for automatic fetal head-pubic symphysis segmentation and AoP measurement. The MiTU-Net model is based on an encoder-decoder framework, utilizing a pre-trained efficient transformer to enhance feature representation. Within the efficient transformer encoder, the model significantly reduces the trainable parameters of the encoder-decoder model. The effectiveness of the proposed method is demonstrated through experiments conducted on a recent transperineal ultrasound dataset. Our model achieves competitive performance, ranking 5th compared to existing approaches. The MiTU-Net presents an efficient method for automatic segmentation and AoP measurement, reducing errors and assisting sonographers in clinical practice. Reproducibility: Framework implementation and models available on https://github.com/13204942/MiTU-Net.
The convolutional neural network-based methods have become more and more popular for medical image segmentation due to their outstanding performance. However, they struggle with capturing long-range dependencies, which are essential for accurately modeling global contextual correlations. Thanks to the ability to model long-range dependencies by expanding the receptive field, the transformer-based methods have gained prominence. Inspired by this, we propose an advanced 2D feature extraction method by combining the convolutional neural network and Transformer architectures. More specifically, we introduce a parallelized encoder structure, where one branch uses ResNet to extract local information from images, while the other branch uses Transformer to extract global information. Furthermore, we integrate pyramid structures into the Transformer to extract global information at varying resolutions, especially in intensive prediction tasks. To efficiently utilize the different information in the parallelized encoder at the decoder stage, we use a channel attention module to merge the features of the encoder and propagate them through skip connections and bottlenecks. Intensive numerical experiments are performed on both aortic vessel tree, cardiac, and multi-organ datasets. By comparing with state-of-the-art medical image segmentation methods, our method is shown with better segmentation accuracy, especially on small organs. The code is publicly available on https://github.com/HongkunSun/ParaTransCNN.
Image restoration tasks traditionally rely on convolutional neural networks. However, given the local nature of the convolutional operator, they struggle to capture global information. The promise of attention mechanisms in Transformers is to circumvent this problem, but it comes at the cost of intensive computational overhead. Many recent studies in image restoration have focused on solving the challenge of balancing performance and computational cost via Transformer variants. In this paper, we present CascadedGaze Network (CGNet), an encoder-decoder architecture that employs Global Context Extractor (GCE), a novel and efficient way to capture global information for image restoration. The GCE module leverages small kernels across convolutional layers to learn global dependencies, without requiring self-attention. Extensive experimental results show that our computationally efficient approach performs competitively to a range of state-of-the-art methods on synthetic image denoising and single image deblurring tasks, and pushes the performance boundary further on the real image denoising task.
Multi-agent motion prediction is a crucial concern in autonomous driving, yet it remains a challenge owing to the ambiguous intentions of dynamic agents and their intricate interactions. Existing studies have attempted to capture interactions between road entities by using the definite data in history timesteps, as future information is not available and involves high uncertainty. However, without sufficient guidance for capturing future states of interacting agents, they frequently produce unrealistic trajectory overlaps. In this work, we propose Future Interaction modeling for Motion Prediction (FIMP), which captures potential future interactions in an end-to-end manner. FIMP adopts a future decoder that implicitly extracts the potential future information in an intermediate feature-level, and identifies the interacting entity pairs through future affinity learning and top-k filtering strategy. Experiments show that our future interaction modeling improves the performance remarkably, leading to superior performance on the Argoverse motion forecasting benchmark.