Memory Generation-Consolidation-Loss and Alzheimer's

Artificial Brain Based on Memory Generation-Consolidation-Loss, Mechanism, Interpretability, Findings and Hypotheses, includes Possible Mechanism of Alzheimer's (Second edition)

As early as 2020, we began to study synaptic strength rebalance, and in October 2021 I had finished the simulation and the paper in Chinese, and also translated most of the Chinese into English, and wrote an Email to Editor-in-Chief of a cell family journal for pre-submission consultation, and he welcomed.

Fig.1 Email

Fig.2 Preprint

Fig.3 Artificial brain flowchart

Fig.4 Memory generation

Fig.5 Memory Consolidation

Fig.6 Memory Loss

Please pay attention our new version of preprint https://arxiv.org/abs/2203.11740, AI+ Brain science +Quantum mechanics+Brain dynamics. We had proposed PNN, but it is not just simple time series predictive models. The cognition of the human brain should be the BP neural network, this cognitive process obtains information from external input to internal output processing information.

In addition to the shared weights of the synaptic connections, we proposed a new neural network that includes the synaptic effective range weights for both the Forward and Back propagation. And lots of simulations were used which RNN cannot be achieved [14-27].

The brain plasticity in positive or negative memory through cognition may be quantum and produce short-term memory, and exhibits an exponential decay in the wave function over a period of time, produced in the hippocampus. And exponential decay occurs due to barriers, and barriers can refer to astrocytes. Brain plasticity in working memory flows through the brain, from the hippocampus to the cortex, through directional derivatives. The strong working memory brain plasticity turns to long-term memory means maximum of directional derivatives, and maximum of directional derivatives is gradient. Thus, long-term memory signifies the gradient of brain plasticity in working memory. The process of short-term memory turns to long-term memory is the process of non-classically turns to classically.

formula (2) and (4), the memory consolidation formula (5) can be shown as follows, it is the relationship of working memory and long-term memory, and maximum of directional derivatives is gradient. From hippocampus to cortices, it is achieved from non-classical to classical in brain. Long-term memory is gradient of brain plasticity

which is too idealistic, there is an angle

,

will increase because of human aging, the angle

is the critical value, greater than it is turbulence and less than it is laminar flow. And turbulence becomes laminar flow gradually throughout aging, the

was shown in Fig. 5 [28]. If

, symptoms of Alzheimer's will appear, and atrophy and hardening of the hippocampus because of reverse turbulence, was shown in Fig. 6, just like a star dying to form black hole. Our conjecture is Alzheimer's cognitive impairment is caused by search direction reversal, the negative gradient of Back propagation is modified by positive gradient, and unable to converge, change of direction of transition was shown in Fig. 5-6 [28]. The

in formula (5) is equivalent to

or

or

in formula (2). The simultaneous interaction of two different turbulence directions of

memory loss and

memory consolidation leads to β-amyloid plaques in brain. But the positive gradient of BP might be the more important reason for Alzheimer's than β-amyloid plaques in brain.

One conjecture about quantum mechanics:

When brain is feeling positive or negative at some time points (it can be seen at the locations of salient points or concave points in Fig.4). The mapping function of exacted memory brain plasticity at these time points may be quantum. And these exacted positive or negative memories brains plasticity will exhibit exponential decay of wave function for a while because of barriers, and exponential decay may be related to human aging. The process of aging may be collection of discontinuous piecewise wave functions of exponential decay. Barriers may relate to astrocytes. Directional derivative of working memory will flow to the different cortices and stored in memory engram cells. Maximum of directional derivatives of working memory means gradient of short-term memory turns to long-term memory.

Negative or positive emotion and cognition reflect non-classical memories by quantum computing, are short-term hippocampal memories. The wave function is high-frequency. And wave function of exacted memories brains plasticity shows kinetic energy because of high-speed particles, produce at hippocampus. By PNN, we propose these appeared barriers from hippocampus to different cortexes, will lead to exponential decay of wave function, and wave function of high-frequency turns to low-frequency. At last, long-term stimulus information stored in memory engram cells of different cortexes. And long-term memory exhibits potential energy release. Barriers may relate to astrocytes. Directional derivative of working memory will flow to the different cortices and stored in memory engram cells. High flow working memory or maximum of directional derivatives of working memory means gradient of short-term memory will turn to long-term memory. Memory flow can take a critical angle of 0, which is the maximum directional derivatives, or a critical angle of

, greater than

memory will flow.

We can imagine our brain as the earth, and the production of geocentric lava occurs like short-term memory happen in the hippocampus, and the process is quantum. Earthquakes on the surface are released due to potential energy, just as strong short-term memory is selected and turns to long-term memory, and is stored in memory engram cells of different cortices, can be released. The Alzheimer's may be the process of planet death, which has the potential to form black holes.

The simulations of PNN fit very well in brain science experiments and hypotheses of 6 papers CNS Journals, 6 papers of CNS family Journals and 2 papers top Physics Journal [14-27].

1. The synaptic strength rebalance, these neurons, like a school of fish, presynaptic neurons like head fishes also affect the locations of postsynaptic neurons [14];

2. The synapse formation causes decline in the number of neurons and impairs brain cognition, then leads to brain aging [15];

3. And the memory of memory engram cells ensembles by a retrograde mechanism, the formula is derived [16];

4. The hippocampal neurogenesis will decline throughout aging [17];

5. But controversy was claimed that human hippocampal neurogenesis persists throughout aging, PNN considered it may have a new and longer circuit in late iteration [18];

6. Closing the critical period which includes astrocytic cortex memory persistence or astrocytes phagocytose synapses at the same time will cause neurological disorder [19];

7. The negative memory will increase activity of brain plasticity [20];

8. Astrocytes phagocytose synapses also inhibits local synaptic accumulation and excitation [21];

9. Relationships of cortex thickness, brain individual diversity and human intelligence [22];

10. The memory retrieval process by memory engram cells that strengthened synaptic strength, increase or decrease synaptic strength [23].

11. Relationship of memory structure and penetrability of brain signals, it means signals go through easily neighboring areas of focus on convex or concave lens. [24];

12．Our brain may be a quantum computer [25], simulation of quantum computer will consider emotion and cognition, when cognition leads to positive or negative emotion, the mapping function of exacted memory brain plasticity may be a non-classical quantum mechanics;

13. Anteromedial thalamus selects strong short-term memories and selectively stabilize memories at remote time [26]. The gradient method with memory gradient to update the synaptic effective range weights is long-term memory stored in cortices. The quantum computing to update the synaptic effective range weights is short-term memory happen in hippocampus. Short-term memory travels through the berries of hippocampus and different cortices and turns into long-term memory. Refer to formula (2), (4) and (5), working memory or short-term memory is consolidated and turns long-term memory. And we suggest strong and high flow working memory or short-term memory means maximum of directional derivatives of exacted memory brain plasticity is relatively good or inferior gradient of memory brain plasticity, means long-term memory;

14. As recently shown in functional magnetic resonance imaging (FMRI) with high spatial resolution, turbulence shows to offer a based way to facilitate energy and information transfer across spatiotemporal scales in brain dynamics [27]. Memory from hippocampus to different cortices because of turbulence rather than laminar flow. Long-term memory is gradient of brain plasticity

that is too idealistic, there is an angle

,

will increase because of human aging, and turbulence becomes laminar flow gradually throughout aging, and smaller value tangent vector

might impair to human intelligence.

About PNN's 4 findings in brain science:

1.Astrocytic cortex memory persistence factor also inhibits local synaptic accumulation and excitation, and the model inspires experiments;

2. It may be the process of astrocytes phagocytose synapses is driven by positive and negative memories of brain plasticity, because of the positive or negative value of

or

;

3. The thicker cortex and the more diverse individuals in brain may have high IQ in simulation, but the thickest cortex and the most diverse individuals in brain may have low IQ in simulation;

4. For PNN, the role of long-term memory is more pronounced than short-term memory.

The innovations of PNN in Deep Learning and Evolutionary Computing:

1.PNN modifies the RNN architecture to be somewhat similar to CNN, and the algorithm is somewhat similar to ResNet, the pooling process or layer number calculation is somewhat similar to update of the synaptic effective range change, PNN also has shared weights of synaptic connections. In addition to the shared weights of synaptic connections, we proposed a new neural network that includes weights of synaptic ranges for Forward propagation and Back propagation;

2.And PNN modified ResNet to calculate the layers’ number, is proposed the gradient of the weight is considered not only the current gradient but also the memory gradient by formula derivation;

3.Such as GA and PSO, they consider global solution or best previous solution, but PNN also considers relatively good solution and relatively inferior solution.

The 4 hypotheses were as follows:

1.Negative or positive emotion and cognition reflect non-classical exacted memories brains plasticity by quantum mechanics, are short-term hippocampal memories. The wave function is high-frequency. And wave function of exacted memories brains plasticity shows kinetic energy because of high-speed particles, produce at hippocampus. By PNN, we propose these appeared barriers from hippocampus to different cortexes, wave function will lead to exponential decay, and wave function of high-frequency turns to low-frequency. At last, long-term stimulus information stored in memory engram cells of different cortexes. And long-term memories exhibit potential energy;

2.Barriers may relate to astrocytes;

3.Directional derivative of working memory will flow to the different cortices and stored in memory engram cells. High flow working memory or maximum of directional derivatives of working memory means gradient of short-term memory, will turn to long-term memory;

4.In PNN simulations, ensembles of long-term memory produced once each iteration, so it occurs once at strong probability. but retrieval processes of short-term memory happened once by many iterations, and occurrences are poor probability.

In fact, except for inspirations the synapse formation [15], quantum computer of brain [25] and memory consolidation [26], the above works are all relevant researches contrast simulation I found later, including the formula reasoning of the ResNet is also completed independently.

This article was funded by JCBV.

[14] El-Boustani, Sami et al. “Locally coordinated synaptic plasticity of visual cortex neurons in vivo.” Science (New York, N.Y.) vol. 360,6395 (2018): 1349-1354. doi:10.1126/science.aao0862

[15] Yukari H. Takeo et al. GluD2- and Cbln1-mediated competitive interactions shape the dendritic arbors of cerebellar Purkinje cells. Neuron, 2021, doi:10.1016/j.neuron.2020.11.028.

[16] Lavi A, Sehgal M, de Sousa AF, Ter-Mkrtchyan D, Sisan F, Luchetti A, Okabe A, Bear C, Silva AJ. Local memory allocation recruits memory ensembles across brain regions. Neuron. 2022 Dec 15:S0896-6273(22)01072-8. doi: 10.1016/j.neuron.2022.11.018. Epub ahead of print. PMID: 36563678.

[17] Sorrells SF, Paredes MF, Cebrian-Silla A, Sandoval K, Qi D, Kelley KW, James D, Mayer S, Chang J, Auguste KI, Chang EF, Gutierrez AJ, Kriegstein AR, Mathern GW, Oldham MC, Huang EJ, Garcia-Verdugo JM, Yang Z, Alvarez-Buylla A. Human hippocampal neurogenesis drops sharply in children to undetectable levels in adults. Nature. 2018 Mar 15;555(7696):377-381. doi: 10.1038/nature25975. Epub 2018 Mar 7. PMID: 29513649; PMCID: PMC6179355.

[18] Boldrini M, Fulmore CA, Tartt AN, Simeon LR, Pavlova I, Poposka V, Rosoklija GB, Stankov A, Arango V, Dwork AJ, Hen R, Mann JJ. Human Hippocampal Neurogenesis Persists throughout Aging. Cell Stem Cell. 2018 Apr 5;22(4):589-599.e5. doi: 10.1016/j.stem.2018.03.015. PMID: 29625071; PMCID: PMC5957089.

[19] Ribot J, Breton R, Calvo CF, Moulard J, Ezan P, Zapata J, Samama K, Moreau M, Bemelmans AP, Sabatet V, Dingli F, Loew D, Milleret C, Billuart P, Dallérac G, Rouach N. Astrocytes close the mouse critical period for visual plasticity. Science. 2021 Jul 2;373(6550):77-81. doi: 10.1126/science.abf5273. PMID: 34210880.

[20] Zhang K, Förster R, He W, Liao X, Li J, Yang C, Qin H, Wang M, Ding R, Li R, Jian T, Wang Y, Zhang J, Yang Z, Jin W, Zhang Y, Qin S, Lu Y, Chen T, Stobart J, Weber B, Adelsberger H, Konnerth A, Chen X. Fear learning induces α7-nicotinic acetylcholine receptor-mediated astrocytic responsiveness that is required for memory persistence. Nat Neurosci. 2021 Dec;24(12):1686-1698. doi: 10.1038/s41593-021-00949-8. Epub 2021 Nov 15. PMID: 34782794.

[21] Lee JH, Kim JY, Noh S, Lee H, Lee SY, Mun JY, Park H, Chung WS. Astrocytes phagocytose adult hippocampal synapses for circuit homeostasis. Nature. 2021 Feb;590(7847):612-617. doi: 10.1038/s41586-020-03060-3. Epub 2020 Dec 23. PMID: 33361813.

[22] Galakhova AA, Hunt S, Wilbers R, Heyer DB, de Kock CPJ, Mansvelder HD, Goriounova NA. Evolution of cortical neurons supporting human cognition. Trends Cogn Sci. 2022 Nov;26(11):909-922. doi: 10.1016/j.tics.2022.08.012. Epub 2022 Sep 15. PMID: 36117080; PMCID: PMC9561064.

[23] Ryan TJ, Roy DS, Pignatelli M, Arons A, Tonegawa S. Memory. Engram cells retain memory under retrograde amnesia. Science. 2015 May 29;348(6238):1007-13. doi: 10.1126/science.aaa5542. Epub 2015 May 28. PMID: 26023136; PMCID: PMC5583719.

[24] Zhao C, Li D, Kong Y, Liu H, Hu Y, Niu H, Jensen O, Li X, Liu H, Song Y. Transcranial photobiomodulation enhances visual working memory capacity in humans. Sci Adv. 2022 Dec 2;8(48):eabq3211. doi: 10.1126/sciadv.abq3211. Epub 2022 Dec 2. PMID: 36459562.

[25] Christian Matthias Kerskens, David López Pérez. Experimental indications of non-classical brain functions. Journal of Physics Communications, 2022, 6(10): 105001. DOI: 10.1088/2399-6528/ac94be.

[26] Toader AC, Regalado JM, Li YR, Terceros A, Yadav N, Kumar S, Satow S, Hollunder F, Bonito-Oliva A, Rajasethupathy P. Anteromedial thalamus gates the selection and stabilization of long-term memories. Cell. 2023 Mar 30;186(7):1369-1381.e17. doi: 10.1016/j.cell.2023.02.024. PMID: 37001501.

[27] Deco, G., Liebana Garcia, S., Sanz Perl, Y. et al. The effect of turbulence in brain dynamics information transfer measured with magnetoencephalography. Commun Phys 6, 74 (2023). https://doi.org/10.1038/s42005-023-01192-2.

[28] Dou, H.-S., Origin of Turbulence-Energy Gradient Theory, 2022, Springer. https://link.springer.com/book/10.1007/978-981-19-0087-7.

完整内容可以阅读原文参考原论文