Direct Preference Optimization (DPO) has emerged as a simple and effective approach for aligning models with human preferences. However, existing DPO-based methods suffer from 3 key drawbacks: they rely on only a single positive-negative preference pair per question, restricting the diversity and richness of feedback; they often emphasize minimizing negative preference scores while neglecting to strengthen the positive preferences; and they depend on either human-annotated preferences or expert model outputs - both expensive and difficult to scale. Moreover, the deterministic ranking assumptions of recent Group-based preference optimization methods break down in open-ended tasks such as Visual Question Answering (VQA), where multiple answers can be equally plausible but differ subtly in relevance or specificity. Given this subtle variance in preferences, we propose to perform ranking over groups of preferences rather than relying on fine-grained ranking of individual ones, which is often noisy and subjective. To address these challenges, we introduce Self-Supervised Visual Preference Alignment via Differentiable Multi-Preference Multi-Group Ranking (SMPRO), a novel framework that (1) self-generates rich, diverse preference groups while eliminating the need for external annotations, (2) employs a fully differentiable ranking objective based on sorting networks to capture nuanced preference gradients across arbitrary numbers of preferences both within and across these groups, and (3) incorporates multiple positive preferences to enrich the positive preference group, capturing subtle distinctions among high-quality preferences. Extensive experiments across diverse visual tasks show that our approach achieves state-of-the-art performance in self-supervised setting. Specifically, our model surpasses existing baselines, achieving notable gains such as 82.4% on MM-Bench, 63.2% on MMStar, 94.6% on LLaVA-W, and 81.9% on AI2D. These results underscore the effectiveness of our approach in capturing richer preference signals and demonstrate its scalability for open-ended, ambiguous VQA tasks.

Recent advances in Multi-Modal Large Language Models (M-LLMs) show promising results in video reasoning. Popular Multi-Modal Large Language Model (M-LLM) frameworks usually apply naive uniform sampling to reduce the number of video frames that are fed into an M-LLM, particularly for long context videos. However, it could lose crucial context in certain periods of a video, so that the downstream M-LLM may not have sufficient visual information to answer a question. To attack this pain point, we propose a light-weight M-LLM-based frame selection method that adaptively select frames that are more relevant to users’ queries. In order to train the proposed frame selector, we introduce two supervision signals (i) Spatial signal, where single frame importance score by prompting an M-LLM; (ii) Temporal signal, in which multiple frames selection by prompting Large Language Model (LLM) using the captions of all frame candidates. The selected frames are then digested by a frozen downstream video M-LLM for visual reasoning and question answering. Empirical results show that the proposed M-LLM video frame selector improves the performances various downstream video Large Language Model (video-LLM) across medium (ActivityNet, NExT-QA) and long (EgoSchema, LongVideoBench) context video question answering benchmarks.

Given a small number of images of a subject, personalized image generation techniques can fine-tune large pretrained text-to-image diffusion models to generate images of the subject in novel contexts, conditioned on text prompts. In doing so, a tradeoff is made between prompt fidelity, subject fidelity and diversity. As the pretrained model is fine-tuned, earlier checkpoints synthesize images with low subject fidelity but high prompt fidelity and diversity. In contrast, later checkpoints generate images with low prompt fidelity and diversity but high subject fidelity. This inherent tradeoff limits the prompt fidelity, subject fidelity and diversity of generated images. In this work, we propose DreamBlend to combine the prompt fidelity from earlier checkpoints and the subject fidelity from later checkpoints during inference. We perform a cross attention guided image synthesis from a later checkpoint, guided by an image

Image to image matching has been well studied in the computer vision community. Previous studies mainly focus on training a deep metric learning model matching visual patterns between the query image and gallery images. In this study, we show that pure image-to- image matching suffers from false positives caused by matching to local visual patterns. To alleviate this issue, we propose to leverage recent advances in vision-language pretraining research. Specifically, we introduce additional image-text alignment losses into deep metric learning, which serve as constraints to the image-to-image matching loss. With additional alignments between the text (e.g., product title) and image pairs, the model can learn concepts from both modalities explicitly, which avoids matching low-level visual features. We progressively develop two variants, a 3-tower and a 4-tower model, where the latter takes one more short text query input. Through extensive experiments, we show that this change leads to a substantial improvement to the image to image matching problem. We further leveraged this model for multimodal search, which takes both image and reformulation text queries to improve search quality. Both offline and online experiments show strong improvements on the main metrics. Specifically, we see 4.95% relative improvement on image matching click through rate with the 3-tower model and 1.13% further improvement from the 4-tower model.

Identifying similar products in e-commerce is useful in discovering relationships between products, making recommendations, and increasing diversity in search results. Product representation learning is the first step to define a generalized product similarity metric for search. The second step is to extend similarity search to a large scale (e.g., e-commerce catalog scale) without sacrificing quality. In this work, we present a solution that interweaves both steps, i.e., learn representations suited to high quality retrieval using contrastive learning (CL) and retrieve similar items from a large search space using approximate nearest neighbor search (ANNS) to trade-off quality for speed. We propose a CL training strategy for learning uni-modal encoders suited to multi-modal similarity search for e-commerce. We study ANNS retrieval by generating Pareto Frontiers (PFs) without requiring labels. Our CL training strategy doubles retrieval@1 metric across categories (e.g., from 36% to 88% in category C). We also demonstrate that ANNS engine optimization using PFs help select configurations appropriately (e.g., we achieve 6.8× search speed with just 2% drop from the maximum retrieval accuracy in medium size datasets).

Despite the prevalence of hypergraphs in a variety of high-impact applications, there are relatively few works on hypergraph representation learning, most of which primarily focus on hyperlink prediction, and are often restricted to the transductive learning setting. Among others, a major hurdle for effective hypergraph representation learning lies in the label scarcity of nodes and/or hyperedges. To address this issue, this paper presents an end-to-end, bi-level pre-training strategy with Graph Neural Networks for hypergraphs. The proposed framework named HyperGRL bears three distinctive advantages. First, it is mainly designed in the self-supervised fashion which has broad applicability, and meanwhile it is also capable of ingesting the labeling information when available. Second, at the heart of the proposed HyperGRL are two carefully designed pretexts, one on the node level and the other on the hyperedge level, which enable us to encode both the local and the global context in a mutually complementary way. Third, the proposed framework can work in both transductive and inductive settings. When applying the two proposed pretexts in tandem, it can accelerate the adaptation of the knowledge from the pre-trained model to downstream applications in the transductive setting, thanks to the bi-level nature of the proposed method. Extensive experiments demonstrate that: (1) HyperGRL achieves up to 5.69% improvements in hyperedge classification, and (2) improves pre-training efficiency by up to 42.80% on average1.

In operant learning, behaviors are reinforced or inhibited in response to the consequences of similar actions taken in the past. However, because in natural environments the "same" situation never recurs, it is essential for the learner to decide what "similar" is so that he can generalize from experience in one state of the world to future actions in different states of the world. The computational principles underlying this generalization are poorly understood, in particular because natural environments are typically too complex to study quantitatively. In this paper we study the principles underlying generalization in operant learning of professional basketball players. In particular, we utilize detailed information about the spatial organization of shot locations to study how players adapt their attacking strategy in real time according to recent events in the game. To quantify this learning, we study how a make\miss from one location in the court affects the probabilities of shooting from different locations. We show that generalization is not a spatially-local process, nor is governed by the difficulty of the shot. Rather, to a first approximation, players use a simplified binary representation of the court into 2 pt and 3 pt zones. This result indicates that rather than using low-level features, generalization is determined by high-level cognitive processes that incorporate the abstract rules of the game.