Peng Su

Aaron Hurst, Adam Lerer, Adam P. Goucher et al. · openai

GPT-4o is an autoregressive omni model that accepts as input any combination of text, audio, image, and video, and generates any combination of text, audio, and image outputs. It's trained end-to-end across text, vision, and audio, meaning all inputs and outputs are processed by the same neural network. GPT-4o can respond to audio inputs in as little as 232 milliseconds, with an average of 320 milliseconds, which is similar to human response time in conversation. It matches GPT-4 Turbo performance on text in English and code, with significant improvement on text in non-English languages, while also being much faster and 50\% cheaper in the API. GPT-4o is especially better at vision and audio understanding compared to existing models. In line with our commitment to building AI safely and consistent with our voluntary commitments to the White House, we are sharing the GPT-4o System Card, which includes our Preparedness Framework evaluations. In this System Card, we provide a detailed look at GPT-4o's capabilities, limitations, and safety evaluations across multiple categories, focusing on speech-to-speech while also evaluating text and image capabilities, and measures we've implemented to ensure the model is safe and aligned. We also include third-party assessments on dangerous capabilities, as well as discussion of potential societal impacts of GPT-4o's text and vision capabilities.

Arpita Vats, Zhe Liu, Peng Su et al.

Model adaptation is crucial to handle the discrepancy between proxy training data and actual users data received. To effectively perform adaptation, textual data of users is typically stored on servers or their local devices, where downstream natural language processing (NLP) models can be directly trained using such in-domain data. However, this might raise privacy and security concerns due to the extra risks of exposing user information to adversaries. Replacing identifying information in textual data with a generic marker has been recently explored. In this work, we leverage large language models (LLMs) to suggest substitutes of masked tokens and have their effectiveness evaluated on downstream language modeling tasks. Specifically, we propose multiple pre-trained and fine-tuned LLM-based approaches and perform empirical studies on various datasets for the comparison of these methods. Experimental results show that models trained on the obfuscation corpora are able to achieve comparable performance with the ones trained on the original data without privacy-preserving token masking.

Shiaho Li, Naisheng Ye, Tianyu Li et al.

Realistic and diverse multi-agent driving scenes are crucial for evaluating autonomous vehicles, but safety-critical events which are essential for this task are rare and underrepresented in driving datasets. Data-driven scene generation offers a low-cost alternative by synthesizing complex traffic behaviors from existing driving logs. However, existing models often lack controllability or yield samples that violate physical or social constraints, limiting their usability. We present OMEGA, an optimization-guided, training-free framework that enforces structural consistency and interaction awareness during diffusion-based sampling from a scene generation model. OMEGA re-anchors each reverse diffusion step via constrained optimization, steering the generation towards physically plausible and behaviorally coherent trajectories. Building on this framework, we formulate ego-attacker interactions as a game-theoretic optimization in the distribution space, approximating Nash equilibria to generate realistic, safety-critical adversarial scenarios. Experiments on nuPlan and Waymo show that OMEGA improves generation realism, consistency, and controllability, increasing the ratio of physically and behaviorally valid scenes from 32.35% to 72.27% for free exploration capabilities, and from 11% to 80% for controllability-focused generation. Our approach can also generate $5\times$ more near-collision frames with a time-to-collision under three seconds while maintaining the overall scene realism.

Szu-Jui Chen, Debjyoti Paul, Yutong Pang et al.

With the emergence of automatic speech recognition (ASR) models, converting the spoken form text (from ASR) to the written form is in urgent need. This inverse text normalization (ITN) problem attracts the attention of researchers from various fields. Recently, several works show that data-driven ITN methods can output high-quality written form text. Due to the scarcity of labeled spoken-written datasets, the studies on non-English data-driven ITN are quite limited. In this work, we propose a language-agnostic data-driven ITN framework to fill this gap. Specifically, we leverage the data augmentation in conjunction with neural machine translated data for low resource languages. Moreover, we design an evaluation method for language agnostic ITN model when only English data is available. Our empirical evaluation shows this language agnostic modeling approach is effective for low resource languages while preserving the performance for high resource languages.

Peng Su, Yifan Peng, K. Vijay-Shanker

Contrastive learning has been used to learn a high-quality representation of the image in computer vision. However, contrastive learning is not widely utilized in natural language processing due to the lack of a general method of data augmentation for text data. In this work, we explore the method of employing contrastive learning to improve the text representation from the BERT model for relation extraction. The key knob of our framework is a unique contrastive pre-training step tailored for the relation extraction tasks by seamlessly integrating linguistic knowledge into the data augmentation. Furthermore, we investigate how large-scale data constructed from the external knowledge bases can enhance the generality of contrastive pre-training of BERT. The experimental results on three relation extraction benchmark datasets demonstrate that our method can improve the BERT model representation and achieve state-of-the-art performance. In addition, we explore the interpretability of models by showing that BERT with contrastive pre-training relies more on rationales for prediction. Our code and data are publicly available at: https://github.com/udel-biotm-lab/BERT-CLRE.

Lei Shi, Kai Shuang, Shijie Geng et al.

Several multi-modality representation learning approaches such as LXMERT and ViLBERT have been proposed recently. Such approaches can achieve superior performance due to the high-level semantic information captured during large-scale multimodal pretraining. However, as ViLBERT and LXMERT adopt visual region regression and classification loss, they often suffer from domain gap and noisy label problems, based on the visual features having been pretrained on the Visual Genome dataset. To overcome these issues, we propose unbiased Contrastive Visual-Linguistic Pretraining (CVLP), which constructs a visual self-supervised loss built upon contrastive learning. We evaluate CVLP on several down-stream tasks, including VQA, GQA and NLVR2 to validate the superiority of contrastive learning on multi-modality representation learning. Our code is available at: https://github.com/ArcherYunDong/CVLP-.

Peng Su, Shudong Huang, Weihong Ma et al.

Previous multi-view contrastive learning methods typically operate at two scales: instance-level and cluster-level. Instance-level approaches construct positive and negative pairs based on sample correspondences, aiming to bring positive pairs closer and push negative pairs further apart in the latent space. Cluster-level methods focus on calculating cluster assignments for samples under each view and maximize view consensus by reducing distribution discrepancies, e.g., minimizing KL divergence or maximizing mutual information. However, these two types of methods either introduce false negatives, leading to reduced model discriminability, or overlook local structures and cannot measure relationships between clusters across views explicitly. To this end, we propose a method named Multi-view Granular-ball Contrastive Clustering (MGBCC). MGBCC segments the sample set into coarse-grained granular balls, and establishes associations between intra-view and cross-view granular balls. These associations are reinforced in a shared latent space, thereby achieving multi-granularity contrastive learning. Granular balls lie between instances and clusters, naturally preserving the local topological structure of the sample set. We conduct extensive experiments to validate the effectiveness of the proposed method.

Tianyu Li, Yihang Qiu, Zhenhua Wu et al.

Multi-traversal data, commonly collected through daily commutes or by self-driving fleets, provides multiple viewpoints for scene reconstruction within a road block. This data offers significant potential for high-quality novel view synthesis, which is crucial for applications such as autonomous vehicle simulators. However, inherent challenges in multi-traversal data often result in suboptimal reconstruction quality, including variations in appearance and the presence of dynamic objects. To address these issues, we propose Multi-Traversal Gaussian Splatting (MTGS), a novel approach that reconstructs high-quality driving scenes from arbitrarily collected multi-traversal data by modeling a shared static geometry while separately handling dynamic elements and appearance variations. Our method employs a multi-traversal dynamic scene graph with a shared static node and traversal-specific dynamic nodes, complemented by color correction nodes with learnable spherical harmonics coefficient residuals. This approach enables high-fidelity novel view synthesis and provides flexibility to navigate any viewpoint. We conduct extensive experiments on a large-scale driving dataset, nuPlan, with multi-traversal data. Our results demonstrate that MTGS improves LPIPS by 23.5% and geometry accuracy by 46.3% compared to single-traversal baselines. The code and data would be available to the public.

Jiamin Xie, Ju Lin, Yiteng Huang et al.

Recent studies have demonstrated that prompting large language models (LLM) with audio encodings enables effective speech recognition capabilities. However, the ability of Speech LLMs to comprehend and process multi-channel audio with spatial cues remains a relatively uninvestigated area of research. In this work, we present directional-SpeechLlama, a novel approach that leverages the microphone array of smart glasses to achieve directional speech recognition, source localization, and bystander cross-talk suppression. To enhance the model's ability to understand directivity, we propose two key techniques: serialized directional output training (S-DOT) and contrastive direction data augmentation (CDDA). Experimental results show that our proposed directional-SpeechLlama effectively captures the relationship between textual cues and spatial audio, yielding strong performance in both speech recognition and source localization tasks.

Shixiang Tang, Peng Su, Dapeng Chen et al.

Human beings can quickly adapt to environmental changes by leveraging learning experience. However, adapting deep neural networks to dynamic environments by machine learning algorithms remains a challenge. To better understand this issue, we study the problem of continual domain adaptation, where the model is presented with a labelled source domain and a sequence of unlabelled target domains. The obstacles in this problem are both domain shift and catastrophic forgetting. We propose Gradient Regularized Contrastive Learning (GRCL) to solve the obstacles. At the core of our method, gradient regularization plays two key roles: (1) enforcing the gradient not to harm the discriminative ability of source features which can, in turn, benefit the adaptation ability of the model to target domains; (2) constraining the gradient not to increase the classification loss on old target domains, which enables the model to preserve the performance on old target domains when adapting to an in-coming target domain. Experiments on Digits, DomainNet and Office-Caltech benchmarks demonstrate the strong performance of our approach when compared to the other state-of-the-art methods.

Peng Su, K. Vijay-Shanker

With the explosive growth of biomedical literature, designing automatic tools to extract information from the literature has great significance in biomedical research. Recently, transformer-based BERT models adapted to the biomedical domain have produced leading results. However, all the existing BERT models for relation classification only utilize partial knowledge from the last layer. In this paper, we will investigate the method of utilizing the entire layer in the fine-tuning process of BERT model. To the best of our knowledge, we are the first to explore this method. The experimental results illustrate that our method improves the BERT model performance and outperforms the state-of-the-art methods on three benchmark datasets for different relation extraction tasks. In addition, further analysis indicates that the key knowledge about the relations can be learned from the last layer of BERT model.

Peng Su, Shixiang Tang, Peng Gao et al.

Human beings can quickly adapt to environmental changes by leveraging learning experience. However, the poor ability of adapting to dynamic environments remains a major challenge for AI models. To better understand this issue, we study the problem of continual domain adaptation, where the model is presented with a labeled source domain and a sequence of unlabeled target domains. There are two major obstacles in this problem: domain shifts and catastrophic forgetting. In this work, we propose Gradient Regularized Contrastive Learning to solve the above obstacles. At the core of our method, gradient regularization plays two key roles: (1) enforces the gradient of contrastive loss not to increase the supervised training loss on the source domain, which maintains the discriminative power of learned features; (2) regularizes the gradient update on the new domain not to increase the classification loss on the old target domains, which enables the model to adapt to an in-coming target domain while preserving the performance of previously observed domains. Hence our method can jointly learn both semantically discriminative and domain-invariant features with labeled source domain and unlabeled target domains. The experiments on Digits, DomainNet and Office-Caltech benchmarks demonstrate the strong performance of our approach when compared to the state-of-the-art.

Peng Su, K. Vijay-Shanker

Adversarial training is a technique of improving model performance by involving adversarial examples in the training process. In this paper, we investigate adversarial training with multiple adversarial examples to benefit the relation extraction task. We also apply adversarial training technique in semi-supervised scenarios to utilize unlabeled data. The evaluation results on protein-protein interaction and protein subcellular localization task illustrate adversarial training provides improvement on the supervised model, and is also effective on involving unlabeled data in the semi-supervised training case. In addition, our method achieves state-of-the-art performance on two benchmarking datasets.

Peng Su, Kun Wang, Xingyu Zeng et al.

Real-world object detectors are often challenged by the domain gaps between different datasets. In this work, we present the Conditional Domain Normalization (CDN) to bridge the domain gap. CDN is designed to encode different domain inputs into a shared latent space, where the features from different domains carry the same domain attribute. To achieve this, we first disentangle the domain-specific attribute out of the semantic features from one domain via a domain embedding module, which learns a domain-vector to characterize the corresponding domain attribute information. Then this domain-vector is used to encode the features from another domain through a conditional normalization, resulting in different domains' features carrying the same domain attribute. We incorporate CDN into various convolution stages of an object detector to adaptively address the domain shifts of different level's representation. In contrast to existing adaptation works that conduct domain confusion learning on semantic features to remove domain-specific factors, CDN aligns different domain distributions by modulating the semantic features of one domain conditioned on the learned domain-vector of another domain. Extensive experiments show that CDN outperforms existing methods remarkably on both real-to-real and synthetic-to-real adaptation benchmarks, including 2D image detection and 3D point cloud detection.

Peng Su, Xiao-Rong Ding, Yuan-Ting Zhang et al.

Existing methods for arterial blood pressure (BP) estimation directly map the input physiological signals to output BP values without explicitly modeling the underlying temporal dependencies in BP dynamics. As a result, these models suffer from accuracy decay over a long time and thus require frequent calibration. In this work, we address this issue by formulating BP estimation as a sequence prediction problem in which both the input and target are temporal sequences. We propose a novel deep recurrent neural network (RNN) consisting of multilayered Long Short-Term Memory (LSTM) networks, which are incorporated with (1) a bidirectional structure to access larger-scale context information of input sequence, and (2) residual connections to allow gradients in deep RNN to propagate more effectively. The proposed deep RNN model was tested on a static BP dataset, and it achieved root mean square error (RMSE) of 3.90 and 2.66 mmHg for systolic BP (SBP) and diastolic BP (DBP) prediction respectively, surpassing the accuracy of traditional BP prediction models. On a multi-day BP dataset, the deep RNN achieved RMSE of 3.84, 5.25, 5.80 and 5.81 mmHg for the 1st day, 2nd day, 4th day and 6th month after the 1st day SBP prediction, and 1.80, 4.78, 5.0, 5.21 mmHg for corresponding DBP prediction, respectively, which outperforms all previous models with notable improvement. The experimental results suggest that modeling the temporal dependencies in BP dynamics significantly improves the long-term BP prediction accuracy.