In this paper, we propose an efficient transformer architecture that uses reinforced positional embedding to obtain superior performance with half the number of encoder decoder layers. We demonstrate that concatenating positional encoding with trainable token embeddings, normalizing columns in the token embedding matrix, and using the normalized token embedding matrix as the value of the attention layer improve the training and validation loss and the training time in an encoder-decoder Transformer model for a Portuguese-English translation task with 10 epochs or 12 hours of training across 10 trials. Our method, with roughly a threefold parameter reduction compared to the baseline model, yields a mean training loss of 1.21, a mean validation loss of 1.51, and an average training time of 1352.27 seconds per epoch, surpassing the baseline model with the same embedding dimension that employs addition of positional encoding and token embeddings, which achieves a mean training loss of 1.96, a validation loss of 2.18, and an average training time of 4297.79 seconds per epoch. Additionally, we evaluated our proposed architecture and the baseline across 14 diverse translation datasets from TensorFlow. The results indicate that our method consistently achieves lower or comparable training and validation losses, suggesting enhanced learning efficiency.
Connectionist Temporal Classification (CTC) is a widely used method for automatic speech recognition (ASR), renowned for its simplicity and computational efficiency. However, it often falls short in recognition performance. In this work, we propose the Consistency-Regularized CTC (CR-CTC), which enforces consistency between two CTC distributions obtained from different augmented views of the input speech mel-spectrogram. We provide in-depth insights into its essential behaviors from three perspectives: 1) it conducts self-distillation between random pairs of sub-models that process different augmented views; 2) it learns contextual representation through masked prediction for positions within time-masked regions, especially when we increase the amount of time masking; 3) it suppresses the extremely peaky CTC distributions, thereby reducing overfitting and improving the generalization ability. Extensive experiments on LibriSpeech, Aishell-1, and GigaSpeech datasets demonstrate the effectiveness of our CR-CTC. It significantly improves the CTC performance, achieving state-of-the-art results comparable to those attained by transducer or systems combining CTC and attention-based encoder-decoder (CTC/AED). We release our code at https://github.com/k2-fsa/icefall.
We initiate a study of the geometry of the visual representation space -- the information channel from the vision encoder to the action decoder -- in an image-based control pipeline learned from behavior cloning. Inspired by the phenomenon of neural collapse (NC) in image classification (arXiv:2008.08186), we empirically demonstrate the prevalent emergence of a similar law of clustering in the visual representation space. Specifically, in discrete image-based control (e.g., Lunar Lander), the visual representations cluster according to the natural discrete action labels; in continuous image-based control (e.g., Planar Pushing and Block Stacking), the clustering emerges according to "control-oriented" classes that are based on (a) the relative pose between the object and the target in the input or (b) the relative pose of the object induced by expert actions in the output. Each of the classes corresponds to one relative pose orthant (REPO). Beyond empirical observation, we show such a law of clustering can be leveraged as an algorithmic tool to improve test-time performance when training a policy with limited expert demonstrations. Particularly, we pretrain the vision encoder using NC as a regularization to encourage control-oriented clustering of the visual features. Surprisingly, such an NC-pretrained vision encoder, when finetuned end-to-end with the action decoder, boosts the test-time performance by 10% to 35%. Real-world vision-based planar pushing experiments confirmed the surprising advantage of control-oriented visual representation pretraining.
Recently, there have been explorations of generalist segmentation models that can effectively tackle a variety of image segmentation tasks within a unified in-context learning framework. However, these methods still struggle with task ambiguity in in-context segmentation, as not all in-context examples can accurately convey the task information. In order to address this issue, we present SINE, a simple image Segmentation framework utilizing in-context examples. Our approach leverages a Transformer encoder-decoder structure, where the encoder provides high-quality image representations, and the decoder is designed to yield multiple task-specific output masks to effectively eliminate task ambiguity. Specifically, we introduce an In-context Interaction module to complement in-context information and produce correlations between the target image and the in-context example and a Matching Transformer that uses fixed matching and a Hungarian algorithm to eliminate differences between different tasks. In addition, we have further perfected the current evaluation system for in-context image segmentation, aiming to facilitate a holistic appraisal of these models. Experiments on various segmentation tasks show the effectiveness of the proposed method.
Video prediction has witnessed the emergence of RNN-based models led by ConvLSTM, and CNN-based models led by SimVP. Following the significant success of ViT, recent works have integrated ViT into both RNN and CNN frameworks, achieving improved performance. While we appreciate these prior approaches, we raise a fundamental question: Is there a simpler yet more effective solution that can eliminate the high computational cost of RNNs while addressing the limited receptive fields and poor generalization of CNNs? How far can it go with a simple pure transformer model for video prediction? In this paper, we propose PredFormer, a framework entirely based on Gated Transformers. We provide a comprehensive analysis of 3D Attention in the context of video prediction. Extensive experiments demonstrate that PredFormer delivers state-of-the-art performance across four standard benchmarks. The significant improvements in both accuracy and efficiency highlight the potential of PredFormer as a strong baseline for real-world video prediction applications. The source code and trained models will be released at https://github.com/yyyujintang/PredFormer.
We introduce UniMuMo, a unified multimodal model capable of taking arbitrary text, music, and motion data as input conditions to generate outputs across all three modalities. To address the lack of time-synchronized data, we align unpaired music and motion data based on rhythmic patterns to leverage existing large-scale music-only and motion-only datasets. By converting music, motion, and text into token-based representation, our model bridges these modalities through a unified encoder-decoder transformer architecture. To support multiple generation tasks within a single framework, we introduce several architectural improvements. We propose encoding motion with a music codebook, mapping motion into the same feature space as music. We introduce a music-motion parallel generation scheme that unifies all music and motion generation tasks into a single transformer decoder architecture with a single training task of music-motion joint generation. Moreover, the model is designed by fine-tuning existing pre-trained single-modality models, significantly reducing computational demands. Extensive experiments demonstrate that UniMuMo achieves competitive results on all unidirectional generation benchmarks across music, motion, and text modalities. Quantitative results are available in the \href{https://hanyangclarence.github.io/unimumo_demo/}{project page}.
Despite the great success of Large Vision-Language Models (LVLMs), they inevitably suffer from hallucination. As we know, both the visual encoder and the Large Language Model (LLM) decoder in LVLMs are Transformer-based, allowing the model to extract visual information and generate text outputs via attention mechanisms. We find that the attention distribution of LLM decoder on image tokens is highly consistent with the visual encoder and both distributions tend to focus on particular background tokens rather than the referred objects in the image. We attribute to the unexpected attention distribution to an inherent flaw in the visual encoder itself, which misguides LLMs to over emphasize the redundant information and generate object hallucination. To address the issue, we propose DAMRO, a novel training-free strategy that $D$ive into $A$ttention $M$echanism of LVLM to $R$educe $O$bject Hallucination. Specifically, our approach employs classification token (CLS) of ViT to filter out high-attention outlier tokens scattered in the background and then eliminate their influence during decoding stage. We evaluate our method on LVLMs including LLaVA-1.5, LLaVA-NeXT and InstructBLIP, using various benchmarks such as POPE, CHAIR, MME and GPT-4V Aided Evaluation. The results demonstrate that our approach significantly reduces the impact of these outlier tokens, thus effectively alleviating the hallucination of LVLMs. The code of our method will be released soon.
Whole slide image (WSI) classification is a crucial problem for cancer diagnostics in clinics and hospitals. A WSI, acquired at gigapixel size, is commonly tiled into patches and processed by multiple-instance learning (MIL) models. Previous MIL-based models designed for this problem have only been evaluated on individual tasks for specific organs, and the ability to handle multiple tasks within a single model has not been investigated. In this study, we propose MECFormer, a generative Transformer-based model designed to handle multiple tasks within one model. To leverage the power of learning multiple tasks simultaneously and to enhance the model's effectiveness in focusing on each individual task, we introduce an Expert Consultation Network, a projection layer placed at the beginning of the Transformer-based model. Additionally, to enable flexible classification, autoregressive decoding is incorporated by a language decoder for WSI classification. Through extensive experiments on five datasets involving four different organs, one cancer classification task, and four cancer subtyping tasks, MECFormer demonstrates superior performance compared to individual state-of-the-art multiple-instance learning models.
U-Net is widely used in medical image segmentation due to its simple and flexible architecture design. To address the challenges of scale and complexity in medical tasks, several variants of U-Net have been proposed. In particular, methods based on Vision Transformer (ViT), represented by Swin UNETR, have gained widespread attention in recent years. However, these improvements often focus on the encoder, overlooking the crucial role of the decoder in optimizing segmentation details. This design imbalance limits the potential for further enhancing segmentation performance. To address this issue, we analyze the roles of various decoder components, including upsampling method, skip connection, and feature extraction module, as well as the shortcomings of existing methods. Consequently, we propose Swin DER (i.e., Swin UNETR Decoder Enhanced and Refined) by specifically optimizing the design of these three components. Swin DER performs upsampling using learnable interpolation algorithm called offset coordinate neighborhood weighted up sampling (Onsampling) and replaces traditional skip connection with spatial-channel parallel attention gate (SCP AG). Additionally, Swin DER introduces deformable convolution along with attention mechanism in the feature extraction module of the decoder. Our model design achieves excellent results, surpassing other state-of-the-art methods on both the Synapse and the MSD brain tumor segmentation task. Code is available at: https://github.com/WillBeanYang/Swin-DER
Large language models (LLMs) have achieved significant performance gains using advanced prompting techniques over various tasks. However, the increasing length of prompts leads to high computational costs and often obscures crucial information. Prompt compression has been proposed to alleviate these issues, but it faces challenges in (i) capturing the global context and (ii) training the compressor effectively. To tackle these challenges, we introduce a novel prompt compression method, namely Reading To Compressing (R2C), utilizing the Fusion-in-Decoder (FiD) architecture to identify the important information in the prompt. Specifically, the cross-attention scores of the FiD are used to discern essential chunks and sentences from the prompt. R2C effectively captures the global context without compromising semantic consistency while detouring the necessity of pseudo-labels for training the compressor. Empirical results show that R2C retains key contexts, enhancing the LLM performance by 6% in out-of-domain evaluations while reducing the prompt length by 80%.
In generative modeling, tokenization simplifies complex data into compact, structured representations, creating a more efficient, learnable space. For high-dimensional visual data, it reduces redundancy and emphasizes key features for high-quality generation. Current visual tokenization methods rely on a traditional autoencoder framework, where the encoder compresses data into latent representations, and the decoder reconstructs the original input. In this work, we offer a new perspective by proposing denoising as decoding, shifting from single-step reconstruction to iterative refinement. Specifically, we replace the decoder with a diffusion process that iteratively refines noise to recover the original image, guided by the latents provided by the encoder. We evaluate our approach by assessing both reconstruction (rFID) and generation quality (FID), comparing it to state-of-the-art autoencoding approaches. By adopting iterative reconstruction through diffusion, our autoencoder, namely Epsilon-VAE, achieves high reconstruction quality, which in turn enhances downstream generation quality by 22% at the same compression rates or provides 2.3x inference speedup through increasing compression rates. We hope this work offers new insights into integrating iterative generation and autoencoding for improved compression and generation.
This technical report introduces a Named Clinical Entity Recognition Benchmark for evaluating language models in healthcare, addressing the crucial natural language processing (NLP) task of extracting structured information from clinical narratives to support applications like automated coding, clinical trial cohort identification, and clinical decision support. The leaderboard provides a standardized platform for assessing diverse language models, including encoder and decoder architectures, on their ability to identify and classify clinical entities across multiple medical domains. A curated collection of openly available clinical datasets is utilized, encompassing entities such as diseases, symptoms, medications, procedures, and laboratory measurements. Importantly, these entities are standardized according to the Observational Medical Outcomes Partnership (OMOP) Common Data Model, ensuring consistency and interoperability across different healthcare systems and datasets, and a comprehensive evaluation of model performance. Performance of models is primarily assessed using the F1-score, and it is complemented by various assessment modes to provide comprehensive insights into model performance. The report also includes a brief analysis of models evaluated to date, highlighting observed trends and limitations. By establishing this benchmarking framework, the leaderboard aims to promote transparency, facilitate comparative analyses, and drive innovation in clinical entity recognition tasks, addressing the need for robust evaluation methods in healthcare NLP.
Transformer-based language models have shown an excellent ability to effectively capture and utilize contextual information. Although various analysis techniques have been used to quantify and trace the contribution of single contextual cues to a target task such as subject-verb agreement or coreference resolution, scenarios in which multiple relevant cues are available in the context remain underexplored. In this paper, we investigate how language models handle gender agreement when multiple gender cue words are present, each capable of independently disambiguating a target gender pronoun. We analyze two widely used Transformer-based models: BERT, an encoder-based, and GPT-2, a decoder-based model. Our analysis employs two complementary approaches: context mixing analysis, which tracks information flow within the model, and a variant of activation patching, which measures the impact of cues on the model's prediction. We find that BERT tends to prioritize the first cue in the context to form both the target word representations and the model's prediction, while GPT-2 relies more on the final cue. Our findings reveal striking differences in how encoder-based and decoder-based models prioritize and use contextual information for their predictions.
Despite their impressive capabilities, multimodal large language models (MLLMs) are prone to hallucinations, i.e., the generated content that is nonsensical or unfaithful to input sources. Unlike in LLMs, hallucinations in MLLMs often stem from the sensitivity of text decoder to visual tokens, leading to a phenomenon akin to "amnesia" about visual information. To address this issue, we propose MemVR, a novel decoding paradigm inspired by common cognition: when the memory of an image seen the moment before is forgotten, people will look at it again for factual answers. Following this principle, we treat visual tokens as supplementary evidence, re-injecting them into the MLLM through Feed Forward Network (FFN) as "key-value memory" at the middle trigger layer. This "look-twice" mechanism occurs when the model exhibits high uncertainty during inference, effectively enhancing factual alignment. Comprehensive experimental evaluations demonstrate that MemVR significantly mitigates hallucination across various MLLMs and excels in general benchmarks without incurring additional time overhead. The implementation is available from https://github.com/1zhou-Wang/MemVR
We propose an Autoregressive (AR) Moving-average (MA) attention structure that can adapt to various linear attention mechanisms, enhancing their ability to capture long-range and local temporal patterns in time series. In this paper, we first demonstrate that, for the time series forecasting (TSF) task, the previously overlooked decoder-only autoregressive Transformer model can achieve results comparable to the best baselines when appropriate tokenization and training methods are applied. Moreover, inspired by the ARMA model from statistics and recent advances in linear attention, we introduce the full ARMA structure into existing autoregressive attention mechanisms. By using an indirect MA weight generation method, we incorporate the MA term while maintaining the time complexity and parameter size of the underlying efficient attention models. We further explore how indirect parameter generation can produce implicit MA weights that align with the modeling requirements for local temporal impacts. Experimental results show that incorporating the ARMA structure consistently improves the performance of various AR attentions on TSF tasks, achieving state-of-the-art results.
Channel coding is vital for reliable data transmission in modern wireless systems, and its significance will increase with the emergence of sixth-generation (6G) networks, which will need to support various error correction codes. However, traditional decoders were typically designed as fixed hardware circuits tailored to specific decoding algorithms, leading to inefficiencies and limited flexibility. To address these challenges, this paper proposes a unified, code-agnostic Transformer-based decoding architecture capable of handling multiple linear block codes, including Polar, Low-Density Parity-Check (LDPC), and Bose-Chaudhuri-Hocquenghem (BCH), within a single framework. To achieve this, standardized units are employed to harmonize parameters across different code types, while the redesigned unified attention module compresses the structural information of various codewords. Additionally, a sparse mask, derived from the sparsity of the parity-check matrix, is introduced to enhance the model's ability to capture inherent constraints between information and parity-check bits, resulting in improved decoding accuracy and robustness. Extensive experimental results demonstrate that the proposed unified Transformer-based decoder not only outperforms existing methods but also provides a flexible, efficient, and high-performance solution for next-generation wireless communication systems.
Capacitive touch sensors capture the two-dimensional spatial profile (referred to as a touch heatmap) of a finger's contact with a mobile touchscreen. However, the research and design of touchscreen mobile keyboards -- one of the most speed and accuracy demanding touch interfaces -- has focused on the location of the touch centroid derived from the touch image heatmap as the input, discarding the rest of the raw spatial signals. In this paper, we investigate whether touch heatmaps can be leveraged to further improve the tap decoding accuracy for mobile touchscreen keyboards. Specifically, we developed and evaluated machine-learning models that interpret user taps by using the centroids and/or the heatmaps as their input and studied the contribution of the heatmaps to model performance. The results show that adding the heatmap into the input feature set led to 21.4% relative reduction of character error rates on average, compared to using the centroid alone. Furthermore, we conducted a live user study with the centroid-based and heatmap-based decoders built into Pixel 6 Pro devices and observed lower error rate, faster typing speed, and higher self-reported satisfaction score based on the heatmap-based decoder than the centroid-based decoder. These findings underline the promise of utilizing touch heatmaps for improving typing experience in mobile keyboards.
We introduce Cadenza, a new multi-stage generative framework for predicting expressive variations of symbolic musical ideas as well as unconditional generations. To accomplish this we propose a novel MIDI encoding method, PerTok (Performance Tokenizer) that captures minute expressive details whilst reducing sequence length up to 59% and vocabulary size up to 95% for polyphonic, monophonic and rhythmic tasks. The proposed framework comprises of two sequential stages: 1) Composer and 2) Performer. The Composer model is a transformer-based Variational Autoencoder (VAE), with Rotary Positional Embeddings (RoPE)ROPE and an autoregressive decoder modified to more effectively integrate the latent codes of the input musical idea. The Performer model is a bidirectional transformer encoder that is separately trained to predict velocities and microtimings on MIDI sequences. Objective and human evaluations demonstrate Cadenza's versatile capability in 1) matching other unconditional state-of-the-art symbolic models in musical quality whilst sounding more expressive, and 2) composing new, expressive ideas that are both stylistically related to the input whilst providing novel ideas to the user. Our framework is designed, researched and implemented with the objective of ethically providing inspiration for musicians.
Detecting 3D keypoints with semantic consistency is widely used in many scenarios such as pose estimation, shape registration and robotics. Currently, most unsupervised 3D keypoint detection methods focus on the rigid-body objects. However, when faced with deformable objects, the keypoints they identify do not preserve semantic consistency well. In this paper, we introduce an innovative unsupervised keypoint detector Key-Grid for both the rigid-body and deformable objects, which is an autoencoder framework. The encoder predicts keypoints and the decoder utilizes the generated keypoints to reconstruct the objects. Unlike previous work, we leverage the identified keypoint in formation to form a 3D grid feature heatmap called grid heatmap, which is used in the decoder section. Grid heatmap is a novel concept that represents the latent variables for grid points sampled uniformly in the 3D cubic space, where these variables are the shortest distance between the grid points and the skeleton connected by keypoint pairs. Meanwhile, we incorporate the information from each layer of the encoder into the decoder section. We conduct an extensive evaluation of Key-Grid on a list of benchmark datasets. Key-Grid achieves the state-of-the-art performance on the semantic consistency and position accuracy of keypoints. Moreover, we demonstrate the robustness of Key-Grid to noise and downsampling. In addition, we achieve SE-(3) invariance of keypoints though generalizing Key-Grid to a SE(3)-invariant backbone.
Arabic handwritten text recognition (HTR) is challenging, especially for historical texts, due to diverse writing styles and the intrinsic features of Arabic script. Additionally, Arabic handwriting datasets are smaller compared to English ones, making it difficult to train generalizable Arabic HTR models. To address these challenges, we propose HATFormer, a transformer-based encoder-decoder architecture that builds on a state-of-the-art English HTR model. By leveraging the transformer's attention mechanism, HATFormer captures spatial contextual information to address the intrinsic challenges of Arabic script through differentiating cursive characters, decomposing visual representations, and identifying diacritics. Our customization to historical handwritten Arabic includes an image processor for effective ViT information preprocessing, a text tokenizer for compact Arabic text representation, and a training pipeline that accounts for a limited amount of historic Arabic handwriting data. HATFormer achieves a character error rate (CER) of 8.6% on the largest public historical handwritten Arabic dataset, with a 51% improvement over the best baseline in the literature. HATFormer also attains a comparable CER of 4.2% on the largest private non-historical dataset. Our work demonstrates the feasibility of adapting an English HTR method to a low-resource language with complex, language-specific challenges, contributing to advancements in document digitization, information retrieval, and cultural preservation.