Title: Partially Observed Structural Causal Models. 

Abstract: Here we introduce Partially Observed Structural Causal Models (POSCMs) as an extension of structural causal models (SCMs) to settings where upstream contexts co-determine both the interaction structure and downstream mechanisms on observed variables. POSCMs thus provide a self-contained causal modeling framework for endogenous graphs, allowing for an intervention hierarchy spanning node- and edge-level contexts and endogenous variable interventions. To define edge interventions, we separate node mechanisms into edge-local transmission channels that can be modified without changing the source node or the rest of the target mechanism. We provide an identifiability theory that clarifies which intervention families would suffice to disentangle structure formation from mechanisms. We then empirically validate these theoretical results in two external simulators: a biophysically detailed virtual human retina and a gene-regulatory analogue. The experiments reproduce non-identifiability under latent context, expose structure-mechanism confounding under latent edges, and recover pathway-level input-output relationships under targeted interventions, consistent with our positive Markov kernel identifiability results. Together, POSCMs provide an intervention-oriented framework for causal systems in which contexts, graph structure, mechanisms, and measurements are jointly generated and only partially observed.

Title: Neural Object Binding: Disentangling What and Where for Robust Visual Tracking and Reasoning.

Abstract: Human visual perception effortlessly parses complex scenes into discrete objects, tracks them through occlusions, and selectively allocates attention—capabilities that remain challenging for artificial vision systems. Inspired by the functional separation of visual processing in the primate brain into ventral (“what”) and dorsal (“where”) streams, this dissertation presents a progressive series of neural architectures that advance object-centric visual understanding. The first contribution, Loci, introduces explicit disentanglement of object identity from spatial position through slot-based representations, achieving state-of-the-art performance on object tracking benchmarks while demonstrating emergent object permanence without explicit supervision. Building on this foundation, Loci-Looped incorporates adaptive perceptual gating that dynamically fuses visual observations with internal predictions, enabling robust tracking through occlusions and exhibiting surprise responses to expectation violations—mirroring developmental findings in infant cognition. Loci-Segmented extends these capabilities to realistic environments through depth-informed spa- tial decomposition and strategic pretraining, achieving competitive segmentation on complex video datasets. The final contribution, FLIP (Fovea-Like Input Patching), addresses computational effi ciency by abandoning whole-image processing in favor of adaptive multi-resolution patch sampling around regions of interest. This biologically inspired selective attention mechanism achieves superior segmentation accuracy compared to state-of-the-art models like the Segment Anything Model (SAM) while using three orders of magnitude fewer parameters and operating twenty-five times faster. Collectively, these architectures demonstrate that some of the key principles of biological vision—representational separation, predictive processing, and selective attention—can be successfully translated into computational systems. The work reveals that intelligent visual processing emerges not from increasing computational capacity, but from learning to allocate existing resources more effectively, suggesting that computational constraints themselves may be fundamental drivers of perceptual intelligence.

Title: How does token structure affect compositional generalization?