And importantly, it wasn't simply about time. In a final experiment, we separated delay from interference.

When they knew which item would be needed, the semantic advantage in non-decision time was almost completely eliminated. When attention remained distributed across several memories, it was large.

Next, we asked what causes this. Using retro-cues, we allowed people to focus attention on one memory before being tested.

Semantic judgements consistently had shorter non-decision times than perceptual judgements, across memory loads and delays. Differences in drift rate were much more conditional, emerging particularly under greater mnemonic demands. Semantic advantage appears primarily before evidence accumulation.

We first revisited our previous working-memory data using hierarchical drift-diffusion modelling. This allowed us to separate two things hidden within a simple reaction time: → processes occurring before evidence accumulation → the accumulation of evidence itself The result was striking.

What happens to a visual memory once it is no longer in front of us? We find that working memory systematically prioritises what something is over what it looked like. But importantly, we wanted to know why.

We also found that hippocampal theta-cortical gamma coupling preceded the increase in cortical dimensionality, suggesting a potential mechanism by which hippocampal output coordinates large-scale cortical reorganisation during retrieval.

We found that successful retrieval was associated with: • stronger cortical reinstatement of encoding-related information • increased cortical representational dimensionality • greater separability of memory representations All emerging shortly after hippocampal ripple onset.

Our central hypothesis was that ripples don't simply reactivate stored memories. Instead, they trigger a transformation from relatively compressed representations into richer, higher-dimensional cortical activity patterns capable of supporting detailed recollection.

8/9 DDM again: sem. effects were primarily pre-accumulation (non-decision time) and selectively larger when items had to be maintained under competing demands.

7/9 Experiment 2 (Immediate vs Delay vs Interference): delay alone didn’t reliably amplify sem. prio, but interference did.

6/9 DDM nailed the dissociation: Valid cueing eliminated the sem. advantage in non-decision time (access demands reduced). But it boosted sem. advantages in drift rate (how efficiently evidence is used once decision starts). So cueing shifts where the sem. edge shows up in the decision pipeline.

5/9 Experiment 1 (retro-cues): tested whether sem. prio is just a decision/selection bias. Valid cue lets you pre-select the to-be-tested item. Neutral cue forces you to keep multiple items available until probe. Result: semantic prioritisation shrinks with valid cues but doesn’t vanish.

4/9 Reanalysis of our earlier dataset: sem. judgements showed a robust reduction in NDT across loads + lags → consistent w faster pre-accumulation access to sem. info. Drift-rate advantages for semantics were conditional (mainly under higher demands: higher load / longer lag btw study & test).

2/9 The striking behavioural pattern: when multiple items are in VWM, people are faster (and often more accurate) for semantic judgements than perceptual ones. That’s the reverse of perception, where low-level visual features typically win the race.

1/9 New paper with @gonzalezgarcia.bsky.social and @lindedomingo.bsky.social : “Characterising semantic prioritisation in visual working memory.” Core question: when we hold visual info briefly in mind, what gets accessed first: perceptual details or semantic meaning?

Critically, inference stretches neural distances along relevant dimensions and compresses irrelevant ones right before a decision, predicts faster RTs, and this re-shaping precedes feedback-related frontal theta tracking model-derived PE.

The brain’s representational space flexes with inferred complexity. Neural effective dimensionality scales up in 2D vs 1D, and is higher on correct vs incorrect trials. In 2D, the two attended features show up as near-orthogonal axes in a shared planar manifold plane. 6/8

Eyes tell the same story Gaze selectively shifts toward task-relevant features, irrelevant features drop out. Gaze entropy decreases as beliefs stabilise, and negative prediction errors from the HSI model trigger broader sampling (exploration), while positive PEs tighten focus (exploitation). 5/8

A Hidden State Inference (HSI) model best explained choices and inferred contexts, beating Q-learning variants (standard, forgetting, counterfactual). HSI captures something structurally different from incremental RL. 4/8

Participants adapted fast: first trial after a switch was at chance level, then rapid recovery. RTs drop and accuracy rises within context blocks - they used the structure to take decisions. 3/8

Serial reversal learning task with same cars, same feature space (3 dimensions), but the rule silently flips. Different dimensions matter in different trials. Sometimes one dim matters, sometimes two dims. You only find out via feedback, meaning participants had to infer the latent state. 2/8

New preprint: Inference over hidden contexts shapes the geometry of conceptual knowledge for flexible behaviour. In this pre-reg study, our core claim was that we don’t just learn stimulus-reward. We infer hidden context and that inference re-wires attention and neural state space on the fly. 1/8

📄 In our new paper, we argue: The best retrieval cue matches the memory now, not just how it was encoded. Always a pleasure working with @lindedomingo.bsky.social. Amusing summary below courtesy of ChatGPT:

🔗 A synchrony bridge? We found theta–gamma phase–amplitude coupling (TG-PAC) between hippocampus and cortex right after ripples. TG-PAC peaks before cortical expansion, suggesting it may help coordinate the shift from compressed hippocampal codes to expanded cortical states.

🌐 After ripples, the brain’s state space unfolds. Cortical dimensionality expands — neural patterns spread apart, making memories easier to decode. More expansion → Faster retrieval and more reinstatement. This raised a question: 🧠 What mechanism is driving this cortical transformation?

🧠 First, ripple characteristics: More ripples on correct vs. incorrect trials 📈 Ripples cluster before memory responses ⏳ Timing suggests ripples help initiate retrieval, not just reflect it.

During an associative memory task, we tracked how ripple events in the hippocampus related to cortical dynamics. 🌟 Hypothesis: Ripples trigger a shift from compressed to expanded neural representations in cortex, making memories readable again.

🧠✨How do we rebuild our memories? In our new study, we show that hippocampal ripples kickstart a coordinated expansion of cortical activity that helps reconstruct past experiences. We recorded iEEG from patients during memory retrieval... and found something really cool 👇(thread)

By examining cross-species evidence, we highlight neural mechanisms that may support episodic memory and identify crucial questions for future research.