Research project · Lecheng Ke ↗
Photos record what we saw. Can we reconstruct what we were?
A research project I have been developing since April 2026, at the edge of memory science, brain–computer interfaces, and ethics. It asks whether future neurotechnology could help someone partially re-approach the inner state of a meaningful moment — and it is built to show exactly where today’s science stops.
- Output
- Thesis, 3D concept map, closed-loop toy simulator
- Evidence
- E1–E2: concept + synthetic data
- Built with
- Python, JavaScript, three.js
01The question
The Unreturnable Present Paradox 当下不可归悖论
You are inside a moment you love, and a quiet thought arrives: I can’t come back here. The grief for the moment starts before the moment ends.
Photos and video keep the evidence of a moment. They don’t keep the state of being in it — the body, the attention, the feeling, the context, the self, and what you didn’t know yet.
What a photo leaves out: an integrated self-state
Seven dimensions, used as a design schema — a conceptual synthesis, not an established neuroscience metric.
- 1Body stateWhat the body predicted it was feeling.
- 2AttentionWhat the mind treated as signal.
- 3EmotionThe shape of the feeling, not a single label.
- 4ContextThe place, people, and story you were inside.
- 5Self-modelWho you felt yourself to be.
- 6MeaningWhy the moment mattered from that life position.
- 7Configured ignoranceWhat you didn’t know yet — which made the moment feel open.
02The idea
Resonance, not replay
Memory is reconstructive — recalling a moment also changes it. So the goal is not to play back an old self. It is a bounded, closed loop that helps the present self re-meet part of a past state, and then come back.
- 1CueA photo, song, place, or smell opens a door to context.
- 2EstimateA coarse guess at the current state — with its uncertainty.
- 3NudgeGentle cueing or neurofeedback, never content “writing”.
- 4MonitorSimilarity, reconstruction error, and emotional risk.
- 5ReturnA gate that always ends in the present.
How far could “going back” go?
Past-state resonance should return you to the present, not trap you in the past.
Design rule · Resonance over Recreation
03Stress test
I tried to break my own idea
Advisor review flagged two structural limits. Instead of hiding them, I built toy experiments that make them visible. Both run on synthetic data: they show the shape of each problem, not results about real brains.
Limit 1 · Decoding power
Can a decoder tell a target memory-state from the others once the signal gets noisy?
Four toy states, six features each, and increasing Gaussian noise (σ). Two classifiers: a nearest-centroid baseline and logistic regression.
Measured points only; the x-axis uses a square-root scale so the low-noise region is readable.
At σ = 0.40, nearest-centroid is at 43.3% and logistic regression at 42.5% (chance is 25%).
Takeaway: in this synthetic experiment, both classifiers approach the 25% chance level as added noise increases and perform similarly across the tested levels. These noise units are not calibrated to EEG.
Limit 2 · Stimulation precision
If we knew where a memory lives, could a grid of electrodes activate it without hitting everything around it?
Amber dots are one memory’s cells; grey dots are everything else. Pick an electrode count and a stimulation radius.
A live re-implementation of the Python sweep (2-D abstraction, not anatomy). Stimulating everything gives ≈ 9% precision — that is the floor.
Challenge: reach ≥ 80% recall with ≥ 20% precision.
Where this leaves the project: today’s technology cannot decode a real past memory or safely stimulate a specific human engram. The honest next step is narrower — test whether cues and feedback can move a person toward a coarse, self-reported state above a null baseline, with uncertainty shown.
Read the full limitations write-up04Go deeper
Explore the project
05Claim boundary
What this is — and isn’t
It is
- A research thesis grounded in adjacent science: reconstructive memory, plasticity, targeted memory reactivation, closed-loop neurofeedback.
- A toy simulation that makes assumptions, uncertainty, and risk visible.
- An honest map of what is known, plausible, and speculative.
It isn’t
- Brain-derived output, memory decoding, or evidence that EEG can read engrams.
- Exact replay, visual playback, or literal time travel.
- A medical device or treatment claim.
- E1Theoretical
- E2Toy scaffold ● current
- E3Validated on real user data
- E4Deployed and tested
- E5Independently replicated