Sculpting dynamical systems for models of neural computation and memory
Bibliographic record
Abstract
Dynamical systems are now a mainstay of computational neuroscience. Persistent activity in biological neural networks has been posited to result from dynamical attractors in neural state-space [1], and computation with attractors underlies a variety of models for information processing and memory function in the brain [2-4]. Nevertheless, there exists no general, substrate-free method for the construction of these dynamical systems [2], substrate-free meaning not tied to a particular neural instantiation. We introduce such a method in this work, with a focus on applications to neural computation and memory modeling. We call this method state-space sculpting. The systems it constructs, or sculpts, are realized as ordinary differential equations. Sculpting is therefore substrate free; it is general in the sense that sculpted systems may contain any number of attractors embedded at any locations in the state-space and the basins of those attractors may be shaped with any smooth geometry. We conjecture that the sculpting process is analogous to learning by synaptic modification in some biological and artificial neural networks. The method works as follows. First, a set of attractor locations and types (fixed point, periodic, strange) is defined. Then a set of corresponding vector fields is identified: Each of these has one of the defined attractor types at one of the defined locations in its state-space. Next, the basins of these attractors are delineated by defining smooth boundaries for the corresponding vector fields. Finally, the set of vector fields and boundaries is unified into a single smooth dynamical equation using regularization [5]. We have sculpted various 2-dimensional systems by this method, and their dynamics were confirmed in simulation to exhibit the defined attractors and basins. These systems contained arbitrary combinations of fixed point and periodic attractors, as well as repelling regions, with arbitrary basin geometries. Systems of saddle equilibria were also sculpted to demonstrate the versatility of the method. These can be made to exhibit heteroclinic channels and cycles, which are key features of the Winnerless Competition Principle of neural computation [4]. As a first test of our conjecture on the analogy between sculpting and synaptic modification, the well known Hopfield network [6] was analyzed and characterized in sculpting terms. This analysis demonstrated most notably that the network's basins of attraction are delineated by combinations of codimension-1 hyperplanes containing saddle equilibria. The analogue of state-space sculpting in the Hopfield network is therefore limited, since attractor basins cannot be shaped with any smooth geometry. Moving forward, it is necessary to study more complicated networks to show that they can be cast in the sculpting framework, as well as to extend the sculpting method to higher-dimensional state-spaces and manifolds.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".