Contrastive Language-Image Pretraining (CLIP) models maximize the mutual information between text and visual modalities to learn representations. This makes the nature of the training data a significant factor in the efficacy of CLIP for downstream tasks. However, the lack of compositional diversity in contemporary image-text datasets limits the compositional reasoning ability of CLIP. We show that generating ``hard'' negative captions via in-context learning and synthesizing corresponding negative images with text-to-image generators offers a solution. We introduce a novel contrastive pre-training strategy that leverages these hard negative captions and images in an alternating fashion to train CLIP. We demonstrate that our method, named TripletCLIP, when applied to existing datasets such as CC3M and CC12M, enhances the compositional capabilities of CLIP, resulting in an absolute improvement of over 9% on the SugarCrepe benchmark on an equal computational budget, as well as improvements in zero-shot image classification and image retrieval. Our code, models, and data are available at: https://tripletclip.github.io

Zero-shot voice conversion (VC) aims to transform the timbre of a source speaker into any previously unseen target speaker, while preserving the original linguistic content. Despite notable progress, attaining a degree of speaker similarity and naturalness on par with ground truth recordings continues to pose great challenge. In this paper, we propose CTEFM-VC, a zero-shot VC framework that leverages Content-aware Timbre Ensemble modeling and Flow Matching. Specifically, CTEFM-VC disentangles utterances into linguistic content and timbre representations, subsequently utilizing a conditional flow matching model and a vocoder to reconstruct the mel-spectrogram and waveform. To enhance its timbre modeling capability and the naturalness of generated speech, we propose a context-aware timbre ensemble modeling approach that adaptively integrates diverse speaker verification embeddings and enables the joint utilization of linguistic and timbre features through a cross-attention module. Experiments show that our CTEFM-VC system surpasses state-of-the-art VC methods in both speaker similarity and naturalness by at least 18.5% and 7.0%.

As modern AI models become integral to everyday tasks, concerns about their inherent biases and their potential impact on human decision-making have emerged. While bias in models are well-documented, less is known about how these biases influence human decisions. This paper presents two interactive experiments investigating the effects of partisan bias in AI language models on political decision-making. Participants interacted freely with either a biased liberal, biased conservative, or unbiased control model while completing political decision-making tasks. We found that participants exposed to politically biased models were significantly more likely to adopt opinions and make decisions aligning with the AI's bias, regardless of their personal political partisanship. However, we also discovered that prior knowledge about AI could lessen the impact of the bias, highlighting the possible importance of AI education for robust bias mitigation. Our findings not only highlight the critical effects of interacting with biased AI and its ability to impact public discourse and political conduct, but also highlights potential techniques for mitigating these risks in the future.

Collaborative Filtering (CF) methods dominate real-world recommender systems given their ability to learn high-quality, sparse ID-embedding tables that effectively capture user preferences. These tables scale linearly with the number of users and items, and are trained to ensure high similarity between embeddings of interacted user-item pairs, while maintaining low similarity for non-interacted pairs. Despite their high performance, encouraging dispersion for non-interacted pairs necessitates expensive regularization (e.g., negative sampling), hurting runtime and scalability. Existing research tends to address these challenges by simplifying the learning process, either by reducing model complexity or sampling data, trading performance for runtime. In this work, we move beyond model-level modifications and study the properties of the embedding tables under different learning strategies. Through theoretical analysis, we find that the singular values of the embedding tables are intrinsically linked to different CF loss functions. These findings are empirically validated on real-world datasets, demonstrating the practical benefits of higher stable rank, a continuous version of matrix rank which encodes the distribution of singular values. Based on these insights, we propose an efficient warm-start strategy that regularizes the stable rank of the user and item embeddings. We show that stable rank regularization during early training phases can promote higher-quality embeddings, resulting in training speed improvements of up to 66%. Additionally, stable rank regularization can act as a proxy for negative sampling, allowing for performance gains of up to 21% over loss functions with small negative sampling ratios. Overall, our analysis unifies current CF methods under a new perspective, their optimization of stable rank, motivating a flexible regularization method.

Synthetic data generation has become an increasingly popular way of training models without the need for large, manually labeled datasets. For tasks like text embedding, synthetic data offers diverse and scalable training examples, significantly reducing the cost of human annotation. However, most current approaches rely heavily on proprietary models like GPT-4, which are expensive and inefficient for generating large-scale embedding data. In this paper, we introduce SPEED, a framework that aligns open-source small models (8B) to efficiently generate large-scale synthetic embedding data. Through supervised fine-tuning, preference optimization, and self-improvement, SPEED enables small open-source models to produce high-quality data. Remarkably, SPEED uses only less than 1/10 of the GPT API calls, outperforming the state-of-the-art embedding model E5_mistral when both are trained solely on their synthetic data. Using this efficient generator, we conduct a comprehensive study on how various factors within the alignment pipeline impact data quality and reveal the scaling law for synthetic embedding data.

Recommender systems are widely used AI applications designed to help users efficiently discover relevant items. The effectiveness of such systems is tied to the satisfaction of both users and providers. However, user satisfaction is complex and cannot be easily framed mathematically using information retrieval and accuracy metrics. While many studies evaluate accuracy through offline tests, a growing number of researchers argue that online evaluation methods such as A/B testing are better suited for this purpose. We have employed a variety of algorithms on different types of datasets divergent in size and subject, producing recommendations in various platforms, including media streaming services, digital publishing websites, e-commerce systems, and news broadcasting networks. Notably, our target websites and datasets are in Persian (Farsi) language. This study provides a comparative analysis of a large-scale recommender system that has been operating for the past year across about 70 websites in Iran, processing roughly 300 requests per second collectively. The system employs user-based and item-based recommendations using content-based, collaborative filtering, trend-based methods, and hybrid approaches. Through both offline and online evaluations, we aim to identify where these algorithms perform most efficiently and determine the best method for our specific needs, considering the dataset and system scale. Our methods of evaluation include manual evaluation, offline tests including accuracy and ranking metrics like hit-rate@k and nDCG, and online tests consisting of click-through rate (CTR). Additionally we analyzed and proposed methods to address cold-start and popularity bias.

Prior study shows that LLMs sometimes generate content that violates copyright. In this paper, we study another important yet underexplored problem, i.e., will LLMs respect copyright information in user input, and behave accordingly? The research problem is critical, as a negative answer would imply that LLMs will become the primary facilitator and accelerator of copyright infringement behavior. We conducted a series of experiments using a diverse set of language models, user prompts, and copyrighted materials, including books, news articles, API documentation, and movie scripts. Our study offers a conservative evaluation of the extent to which language models may infringe upon copyrights when processing user input containing protected material. This research emphasizes the need for further investigation and the importance of ensuring LLMs respect copyright regulations when handling user input to prevent unauthorized use or reproduction of protected content. We also release a benchmark dataset serving as a test bed for evaluating infringement behaviors by LLMs and stress the need for future alignment.

We investigate what kind of images lie in the high-density regions of diffusion models. We introduce a theoretical mode-tracking process capable of pinpointing the exact mode of the denoising distribution, and we propose a practical high-probability sampler that consistently generates images of higher likelihood than usual samplers. Our empirical findings reveal the existence of significantly higher likelihood samples that typical samplers do not produce, often manifesting as cartoon-like drawings or blurry images depending on the noise level. Curiously, these patterns emerge in datasets devoid of such examples. We also present a novel approach to track sample likelihoods in diffusion SDEs, which remarkably incurs no additional computational cost.

With the advent of social media networks and the vast amount of information circulating through them, automatic fact verification is an essential component to prevent the spread of misinformation. It is even more useful to have fact verification systems that provide explanations along with their classifications to ensure accurate predictions. To address both of these requirements, we implement AMREx, an Abstract Meaning Representation (AMR)-based veracity prediction and explanation system for fact verification using a combination of Smatch, an AMR evaluation metric to measure meaning containment and textual similarity, and demonstrate its effectiveness in producing partially explainable justifications using two community standard fact verification datasets, FEVER and AVeriTeC. AMREx surpasses the AVeriTec baseline accuracy showing the effectiveness of our approach for real-world claim verification. It follows an interpretable pipeline and returns an explainable AMR node mapping to clarify the system's veracity predictions when applicable. We further demonstrate that AMREx output can be used to prompt LLMs to generate natural-language explanations using the AMR mappings as a guide to lessen the probability of hallucinations.

Rating is a typical user explicit feedback that visually reflects how much a user likes a related item. The (rating) matrix completion is essentially a rating prediction process, which is also a significant problem in recommender systems. Recently, graph neural networks (GNNs) have been widely used in matrix completion, which captures users' preferences over items by formulating a rating matrix as a bipartite graph. However, existing methods are susceptible due to data sparsity and long-tail distribution in real-world scenarios. Moreover, the messaging mechanism of GNNs makes it difficult to capture high-order correlations and constraints between nodes, which are essentially useful in recommendation tasks. To tackle these challenges, we propose a Multi-Channel Hypergraph Contrastive Learning framework for matrix completion, named MHCL. Specifically, MHCL adaptively learns hypergraph structures to capture high-order correlations between nodes and jointly captures local and global collaborative relationships through attention-based cross-view aggregation. Additionally, to consider the magnitude and order information of ratings, we treat different rating subgraphs as different channels, encourage alignment between adjacent ratings, and further achieve the mutual enhancement between different ratings through multi-channel cross-rating contrastive learning. Extensive experiments on five public datasets demonstrate that the proposed method significantly outperforms the current state-of-the-art approaches.