Researchers introduce MOJO (Masked autOencoder-based JOint training), a training framework that augments spike-tokenizing neural decoding models with self-supervised learning via masked autoencoding alongside supervised objectives. Evaluated on spiking datasets from monkey motor cortex and multi-regional mouse recordings, MOJO outperforms purely supervised models, with especially strong gains in few-shot and label-limited settings. The framework also generalizes to human electrocorticography during speech, achieving performance comparable to neuro-foundation models designed for continuous signals. The work advances a path toward more scalable and flexible training of neural foundation models using unlabelled data.
BrainJanus is a unified model that integrates brain neural activity, vision, and language within a single autoregressive framework using next-token prediction. The system introduces a Unified Brain Tokenizer to quantize neural dynamics into discrete tokens aligned with visual and linguistic representations in a shared embedding space. It supports any-to-any generation including image-to-brain, text-to-brain, brain-to-image, and brain-to-text tasks, with reported zero-shot generalization and interpretable biological topography. The work positions itself as a general-purpose brain modeling paradigm at the intersection of neuroscience and multimodal AI.
Researchers propose converting a self-supervised speech representation model into a Mixture-of-Experts (MoE) architecture to improve generalization in synthetic speech detection. Feed-forward blocks in selected encoder layers are replaced by expert networks with a layer-wise gating mechanism, allowing complementary acoustic pattern capture while preserving pretrained representations. Evaluated across 14 spoofing datasets, the approach reduces macro Equal Error Rate from 5.46% to 4.81%, an 11.9% relative improvement over the baseline.
Researchers propose a masked multimodal speech synthesis framework that jointly trains on surface electromyography (sEMG) and video-based lipreading signals using modality masking to improve robustness to sensor failure or degradation. In multispeaker settings, the approach reduces word error rate by up to 14 absolute percentage points over the strongest unimodal baseline. Masking strategies outperform degradation-specific data augmentation for handling missing modalities, with phone-level analysis revealing complementary contributions across vowels and consonant groups.
Researchers at EPFL introduce Modus, a decoder-only model that treats all modalities symmetrically for any-to-any prediction without modality-specific heads, losses, or task pipelines. Unlike prior any-to-any models that use encoder-decoder or diffusion architectures trained from scratch, Modus leverages pre-trained decoder-only models as a prior. The model supports chained generation through intermediate modalities and cross-modal self-verification, achieving competitive performance with specialist and multitask baselines across benchmarks. All materials are open-sourced.
Meta AI released Brain2Qwerty v2, an end-to-end deep learning pipeline that decodes text from non-invasive magnetoencephalography (MEG) brain recordings in real time, achieving 61% word accuracy — up from 8% for prior non-invasive methods and approaching surgical-implant performance. The system was trained on ~22,000 sentences from nine participants and uses fine-tuned large language models on neural data to bridge noisy brain signals and coherent language. Meta is releasing full training code for both v1 and v2, and partner institution BCBL is releasing the v1 dataset. The work is part of Meta's broader Digital Brain Project and open neuroscience initiative.
AllenAI introduces EMO, a pretraining approach for Mixture of Experts (MoE) models that aims to produce emergent modularity during training. The work explores how MoE architectures can develop specialized expert routing without explicit supervision. Published on the Hugging Face blog, this represents research-level work on improving MoE training dynamics and efficiency.
Researchers propose AGDO, a framework that replaces random masking in diffusion large language models (dLLMs) with attention-guided denoising order and token weighting during fine-tuning and reinforcement learning. The work is motivated by an empirical finding that tokens with stronger attention to unmasked context are more stable and critical for reasoning. Experiments on math and coding benchmarks show AGDO outperforms existing post-training methods for dLLMs, advancing the case for attention-aware training in parallel-decoding language models.
The paper introduces TxFM, a self-supervised masked autoencoder model for transcriptomic (RNA-seq) data representation learning, trained on a curated 1.4M-sample corpus called DiverseRNA-1.4M. TxFM outperforms existing transcriptomic foundation models trained on datasets over 100x larger, addressing the known problem of deep models underperforming linear baselines on gene expression data. The work provides ablation studies identifying critical architecture choices and argues that careful data curation combined with inductive self-supervised learning is sufficient for strong transfer performance in transcriptomics.