Why weakening synapses and strengthening memories can happen at once
At first the two ideas look contradictory. One says sleep broadly reduces synaptic strength, a kind of global reset that keeps the brain from saturating [1][2]. The other says sleep consolidates particular memories by replaying the same activity patterns that occurred during learning, which should strengthen the synapses involved [4]. The reconciliation is that these operate at different scales and on different rules: downscaling is global and non-selective, while replay is selective and targets specific representations [2].
A useful comparison is a library that thins its shelves overnight. Everything gets a little less prominent, but the books that are repeatedly taken out and re-shelved are the ones that remain easy to find. In the brain, the replayed patterns are the books being taken out; the global weakening is the thinning. The result is not that all memories weaken equally, but that replayed circuits become relatively more distinctive against a quieter background [2].
Replay during slow-wave sleep is embedded in the downscaling process
One influential account treats long-term memory formation during sleep as an active systems-consolidation process that is embedded within global synaptic downscaling [2]. During slow-wave sleep, the hippocampus repeatedly reactivates representations, and this replay gradually transforms and integrates them into neocortical networks [2]. The same sleep stage is associated with both hippocampo-cortical replay through sharp-wave ripple events and synaptic downscaling, so the two are described as features of the same period rather than separate competing events [1].
Modeling work offers a mechanistic picture of how weakening and replay can coexist. In a biophysically detailed model of the entorhinal-hippocampal-cortical network, the transition into slow-wave sleep through dampening of inhibition generates up-down states and spontaneous, time-compressed replays of spatial sequences [3]. Weakened afferent synapses to CA1 eliminated ripples but spared the replays themselves, and replays continued to sustain systems consolidation without ripples, though more slowly [3]. That suggests weakening can proceed while the replay signal persists, at least up to a point: excessive weakening halted consolidation, and it could be rescued by enhancing hippocampal-to-cortical connectivity [3].
Potentiation and depression may run in parallel, not in sequence
Another way the conflict softens is that sleep does not appear to be purely a weakening state. Two forms of synaptic plasticity, potentiation and depression of synaptic transmission, are described as being induced in parallel during sleep, collectively called offline synaptic plasticity, as distinct from the online plasticity that occurs right after a memory event [4]. Replay of awake activity patterns during sleep is thought to trigger plastic changes in synaptic transmission across multiple brain regions, and perturbing that replay blocks long-term memory formation [4].
So the picture is not "weaken everything, then strengthen a few." It is closer to a single offline period in which depression and potentiation occur together, with replay biasing which synapses get the potentiating side of the ledger [4]. The global downscaling account and the replay account then describe different aspects of the same offline window rather than incompatible claims about it [2][4].
What remains genuinely open is the accounting. The evidence supports the idea that global downscaling and replay-based consolidation are coordinated during slow-wave sleep [1][2], and modeling shows replay can survive substantial synaptic weakening while consolidation slows and can fail if weakening goes too far [3]. But how the brain balances the two in any given night, and how that balance differs across sleep stages and species, is still being worked out; the rodent findings come largely from animals with polyphasic sleep, which limits direct extrapolation to human monophasic sleep architecture [1].
Sources used in this answer
Neurobiological Mechanisms of Sleep-Dependent Memory Consolidation in Rodent Models: From Neural Oscillations to Synaptic Plasticity and Behavioral Outcomes
Rodent evidence links slow-wave sleep to both hippocampo-cortical replay via sharp-wave ripples and synaptic downscaling, with NREM and REM stages supporting different aspects of memory processing.
Mechanisms of systems memory consolidation during sleep.
Long-term memory formation during sleep is proposed as active systems consolidation embedded in global synaptic downscaling, with repeated hippocampal replay during slow-wave sleep gradually integrating representations into neocortical networks.
Sleep-modulated disinhibition enables replay for memory consolidation, accelerated by ripples
In a biophysically detailed network model, sleep-related disinhibition generates replay, weakened CA1 afferent synapses eliminate ripples but spare replays, and replays sustain consolidation without ripples though more slowly, while excessive weakening halts it.
Offline neuronal activity and synaptic plasticity during sleep and memory consolidation
Replay of awake activity patterns during sleep is thought to trigger synaptic plasticity, with potentiation and depression induced in parallel as offline synaptic plasticity, and replay perturbation blocks long-term memory formation.
Learning to Forget: Sleep-Inspired Memory Consolidation for Resolving Proactive Interference in Large Language Models
A sleep-inspired framework for large language models uses synaptic downscaling, selective replay, and targeted forgetting to reduce proactive interference in a key-value cache.