Text-to-image diffusion models excel at translating language prompts into photorealistic images by implicitly grounding textual concepts through their cross-modal attention mechanisms. Recent multi-modal diffusion transformers extend this by introducing joint self-attention over concatenated image and text tokens, enabling richer and more scalable cross-modal alignment. However, a detailed understanding of how and where these attention maps contribute to image generation remains limited. In this paper, we introduce Seg4Diff (Segmentation for Diffusion), a systematic framework for analyzing the attention structures of MM-DiT, with a focus on how specific layers propagate semantic information from text to image. Through comprehensive analysis, we identify a semantic grounding expert layer, a specific MM-DiT block that consistently aligns text tokens with spatially coherent image regions, naturally producing high-quality semantic segmentation masks. We further demonstrate that applying a lightweight fine-tuning scheme with mask-annotated image data enhances the semantic grouping capabilities of these layers and thereby improves both segmentation performance and generated image fidelity. Our findings demonstrate that semantic grouping is an emergent property of diffusion transformers and can be selectively amplified to advance both segmentation and generation performance, paving the way for unified models that bridge visual perception and generation.

Bilinear feature transformation has shown the state-of-the-art performance in learning fine-grained image representations.

The proposed group bilinear requires the intra-group channels to be highly5 correlated (refer tothedefinitioninQ3.1),andtheproposed semantic grouping canbetter satisfy suchrequirements6 than MA-CNN [9]. Specifically,[9] adopts the idea of k-means, which optimizes each channel to its cluster center.7 Note that the notations aboveare the same with Eqn.16 (3),andthepairwisecorrelationis dij = Thanks foryour comments.Aisanapproximate indexmapping20 matrix, whose rows are constrained to be (approximate) one-hot vectors via asoftmax with small "temperature".21 Q2.2Inconsistentnotations. Thanks for your comments, and we will correct the notation "stage 3,4" into "Stage28 IV,V"respectively. Designing suitable grouping methods plays a42 keyrole.

In this paper, we tackle the problem of learning visual representations from unlabeled scene-centric data. Existing works have demonstrated the potential of utilizing the underlying complex structure within scene-centric data; still, they commonly rely on hand-crafted objectness priors or specialized pretext tasks to build a learning framework, which may harm generalizability. Instead, we propose contrastive learning from data-driven semantic slots, namely SlotCon, for joint semantic grouping and representation learning. The semantic grouping is performed by assigning pixels to a set of learnable prototypes, which can adapt to each sample by attentive pooling over the feature and form new slots. Based on the learned data-dependent slots, a contrastive objective is employed for representation learning, which enhances the discriminability of features, and conversely facilitates grouping semantically coherent pixels together. Compared with previous efforts, by simultaneously optimizing the two coupled objectives of semantic grouping and contrastive learning, our approach bypasses the disadvantages of hand-crafted priors and is able to learn object/group-level representations from scene-centric images. Experiments show our approach effectively decomposes complex scenes into semantic groups for feature learning and significantly benefits downstream tasks, including object detection, instance segmentation, and semantic segmentation.

In this paper, we tackle the problem of learning visual representations from unlabeled scene-centric data. Existing works have demonstrated the potential of utilizing the underlying complex structure within scene-centric data; still, they commonly rely on hand-crafted objectness priors or specialized pretext tasks to build a learning framework, which may harm generalizability. Instead, we propose contrastive learning from data-driven semantic slots, namely SlotCon, for joint semantic grouping and representation learning.

Recent advances in finance-specific language models such as FinBERT have enabled the quantification of public sentiment into index-based measures, yet compressing diverse linguistic signals into single metrics overlooks contextual nuances and limits interpretability. To address this limitation, explainable AI techniques, particularly SHAP (SHapley Additive Explanations), have been employed to identify influential features. However, SHAP's computational cost grows exponentially with input features, making it impractical for large-scale text-based financial data. This study introduces a GRU-based forecasting framework enhanced with GroupSHAP, which quantifies contributions of semantically related keyword groups rather than individual tokens, substantially reducing computational burden while preserving interpretability. We employed FinBERT to embed news articles from 2015 to 2024, clustered them into coherent semantic groups, and applied GroupSHAP to measure each group's contribution to stock price movements. The resulting group-level SHAP variables across multiple topics were used as input features for the prediction model. Empirical results from one-day-ahead forecasting of the S&P 500 index throughout 2024 demonstrate that our approach achieves a 32.2% reduction in MAE and a 40.5% reduction in RMSE compared with benchmark models without the GroupSHAP mechanism. This research presents the first application of GroupSHAP in news-driven financial forecasting, showing that grouped sentiment representations simultaneously enhance interpretability and predictive performance.

Understanding the locus of semantic representation in large language models (LLMs) is crucial for interpretability and architectural innovation. The dominant paradigm posits that trainable input embeddings serve as foundational "meaning vectors." This paper challenges that view. We construct Transformer models where the embedding layer is entirely frozen, with vectors derived not from data, but from the visual structure of Unicode glyphs. These non-semantic, precomputed visual embeddings are fixed throughout training. Our method is compatible with any tokenizer, including a novel Unicode-centric tokenizer we introduce to ensure universal text coverage. Despite the absence of trainable, semantically initialized embeddings, our models converge, generate coherent text, and, critically, outperform architecturally identical models with trainable embeddings on the MMLU reasoning benchmark. We attribute this to "representational interference" in conventional models, where the embedding layer is burdened with learning both structural and semantic features. Our results indicate that high-level semantics are not inherent to input embeddings but are an emergent property of the Transformer's compositional architecture and data scale. This reframes the role of embeddings from meaning containers to structural primitives. We release all code and models to foster further research.

Neural Information Processing Systems http://nips.cc/