Searcharxiv⌕ Search

arXiv subjects

Yusuf Shihata

Publications and source records attributed to Yusuf Shihata.

2 recordsLinked to original sources

Asymmetric Multimodal Learning with Decomposed JEPAs

Cross-modal alignment is the standard objective for building multimodal representations. We show that alignment as a \emph{sole} objective is lossy wherever modalities are informationally asymmetric---where neither can express what the other carries. Visual question answering is an obvious case: a question is a selective relational operator probing a dense visual field whose fine-grained features no query recovers, so what alignment preserves is not what the task requires. We prove an \textbf{Alignment Redundancy Bound}: alignment-only Dirichlet energy minimization retains at most the redundant information atom $\Red(Z; X_1, X_2)$, annihilating private modality-specific structure $\Unq(Z; X_m)$ and cross-modal synergy $\Syn(Z; X_1, X_2)$. We introduce \textbf{Decomposed JEPA (D-JEPA)}, which factorizes latents into an aligned shared core and private modality-specific fibers. The shared core retains most of the structure alignment targets; the private fibers recover what alignment discards, via contextual masked prediction and asymmetric cross-subspace independence. Under information asymmetry, alignment-only training undergoes catastrophic semantic collapse on compositional reasoning benchmarks, surviving only on binary queries. D-JEPA averts this collapse, substantially improving downstream accuracy over both alignment-only and alignment-with-reconstruction baselines, and restoring a positive multimodal synergy gap in place of the baseline's inverted gap. Information-theoretic audits confirm that D-JEPA recovers most of the mutual information available from a frozen backbone, and causal fiber-swap interventions demonstrate that the recovered information is causally localized in the private fibers.

cs.LG↗

Gated Recursive Fusion: A Stateful Approach to Scalable Multimodal Transformers

Multimodal learning faces a fundamental tension between deep, fine-grained fusion and computational scalability. While cross-attention models achieve strong performance through exhaustive pairwise fusion, their quadratic complexity is prohibitive for settings with many modalities. We address this challenge with Gated Recurrent Fusion (GRF), a novel architecture that captures the power of cross-modal attention within a linearly scalable, recurrent pipeline. Our method processes modalities sequentially, updating an evolving multimodal context vector at each step. The core of our approach is a fusion block built on Transformer Decoder layers that performs symmetric cross-attention, mutually enriching the shared context and the incoming modality. This enriched information is then integrated via a Gated Fusion Unit (GFU) a GRU-inspired mechanism that dynamically arbitrates information flow, enabling the model to selectively retain or discard features. This stateful, recurrent design scales linearly with the number of modalities, O(n), making it ideal for high-modality environments. Experiments on the CMU-MOSI benchmark demonstrate that GRF achieves competitive performance compared to more complex baselines. Visualizations of the embedding space further illustrate that GRF creates structured, class-separable representations through its progressive fusion mechanism. Our work presents a robust and efficient paradigm for powerful, scalable multimodal representation learning.

cs.CV↗