Assel Kembay

Rui-Jie Zhu, Tianhao Peng, Tianhao Cheng et al.

Large Language Models (LLMs) have demonstrated impressive reasoning capabilities, especially when guided by explicit chain-of-thought (CoT) reasoning that verbalizes intermediate steps. While CoT improves both interpretability and accuracy, its dependence on natural language reasoning limits the model's expressive bandwidth. Latent reasoning tackles this bottleneck by performing multi-step inference entirely in the model's continuous hidden state, eliminating token-level supervision. To advance latent reasoning research, this survey provides a comprehensive overview of the emerging field of latent reasoning. We begin by examining the foundational role of neural network layers as the computational substrate for reasoning, highlighting how hierarchical representations support complex transformations. Next, we explore diverse latent reasoning methodologies, including activation-based recurrence, hidden state propagation, and fine-tuning strategies that compress or internalize explicit reasoning traces. Finally, we discuss advanced paradigms such as infinite-depth latent reasoning via masked diffusion models, which enable globally consistent and reversible reasoning processes. By unifying these perspectives, we aim to clarify the conceptual landscape of latent reasoning and chart future directions for research at the frontier of LLM cognition. An associated GitHub repository collecting the latest papers and repos is available at: https://github.com/multimodal-art-projection/LatentCoT-Horizon/.

Skye Gunasekaran, Assel Kembay, Hugo Ladret et al.

Accurate time-series forecasting is crucial in various scientific and industrial domains, yet deep learning models often struggle to capture long-term dependencies and adapt to data distribution shifts over time. We introduce Future-Guided Learning, an approach that enhances time-series event forecasting through a dynamic feedback mechanism inspired by predictive coding. Our method involves two models: a detection model that analyzes future data to identify critical events and a forecasting model that predicts these events based on current data. When discrepancies occur between the forecasting and detection models, a more significant update is applied to the forecasting model, effectively minimizing surprise, allowing the forecasting model to dynamically adjust its parameters. We validate our approach on a variety of tasks, demonstrating a 44.8% increase in AUC-ROC for seizure prediction using EEG data, and a 23.4% reduction in MSE for forecasting in nonlinear dynamical systems (outlier excluded).By incorporating a predictive feedback mechanism, Future-Guided Learning advances how deep learning is applied to time-series forecasting.

Yuchen Tian, Assel Kembay, Samuel Tensingh et al.

Spiking neural networks (SNNs) promise energy-efficient computation by mimicking biological neural dynamics, yet existing plasticity rules focus on isolated spike pairs and fail to leverage the synchronous activity patterns that drive learning in biological systems. We introduce spike-synchrony-dependent plasticity (SSDP), a training approach that adjusts synaptic weights based on the degree of synchronous neural firing rather than spike timing order. Our method operates as a local, post-optimization mechanism that applies updates to sparse parameter subsets, maintaining computational efficiency with linear scaling. SSDP serves as a lightweight event-structure regularizer, biasing the network toward biologically plausible spatio-temporal synchrony while preserving standard convergence behavior. SSDP seamlessly integrates with standard backpropagation while preserving the forward computation graph. We validate our approach across single-layer SNNs and spiking Transformers on datasets from static images to high-temporal-resolution tasks, demonstrating improved convergence stability and enhanced robustness to spike-time jitter and event noise. These findings provide new insights into how biological neural networks might leverage synchronous activity for efficient information processing and suggest that synchrony-dependent plasticity represents a key computational principle underlying neural learning.