Get Started
Topic · 2 recaps
Neural and Evolutionary Computing
Neural architectures and training dynamics, plus the evolutionary and biologically-inspired approaches that share this corner of arXiv with mainstream deep learning.
Play all
...
Posts
Questions
Home
Topics
Search
Library
Questions researchers are working on
Follow a question through Rcap’s explanations and the latest papers addressing it.
Search
Can autoencoder parameter spectra serve as usable representations of their training data?
The statistical properties of training data may be reflected in the singular values of an autoencoder’s parameter matrices. The difficulty is determining whether this information remains sufficiently distinctive and usable as a data representation.
Can intermediate LLM activations guide faster jailbreak search without weakening attack effectiveness?
Refusal behavior may be represented in transformer activations before the model produces its output. The difficulty is using that signal to reduce the cost of prompt search without losing the effectiveness of the resulting attacks.
Can non-smooth or quantized activations support stable echo-state dynamics beyond conventional spectral-radius expectations?
Echo state network stability is often analyzed using smooth activations and conservative spectral-radius conditions. Irregular or quantized activations may change how reservoir states contract, remain distinct, or converge, but their stability behavior is not fully understood.
Can pre-generalization interventions reveal when training constrains which equally fitting neural-network solutions will later generalize?
Overparameterized networks can fit the same training data while differing substantially on unseen examples. During grokking, generalization emerges after a plateau, making it difficult to determine when training has begun constraining the eventual solution.
Can stochastic weight averaging improve equivariance in augmented classification without repeated ensemble training?
Data augmentation incorporates task symmetries into neural networks, while deep ensembles can require many separate training runs. The practical difficulty is determining whether weight averaging can provide stronger symmetry handling without that repeated cost.
How can Adam’s error be bounded for strongly convex stochastic optimization without assuming bounded iterates?
Analyses of Adam on strongly convex stochastic problems have often assumed that its iterates remain uniformly bounded. Without that premise, it is unclear whether the optimizer’s error can be controlled unconditionally.
How can agents decide how to combine reusable behaviors as circumstances change?
Systems may have several learned behaviors available, but it is unclear how context should determine their relative contributions while keeping the composition process dynamically coherent. Gating rules, update dynamics, and neural implementations are often treated as separate design problems.
How can an agent substrate support self-reproduction and self-evolution without losing constructional heredity?
An agent that changes its own capabilities must preserve enough constructional information for those changes to remain reproducible in descendants. Combining maintenance, evolution, reproduction, and organization also makes boundaries, identity, transitions, and inheritance interdependent.
How can an autonomous audio system evolve sonic behavior without external data or post-initialization supervision?
Without external data or feedback after initialization, an audio system must both produce changing sound and determine how its internal parameters should change. The difficulty is sustaining autonomous sonic evolution without simply becoming static or unstable.
How can associative-memory capacity be compared across Hopfield and attention-like models without conflating assumptions?
Hopfield-style memories combine recurrent dynamics, energy landscapes, and pattern storage, but retrieval capacity depends on the disorder ensemble, scaling limit, and success criterion. Connections to attention and biological interpretation introduce further assumptions that can make superficially similar results incomparable.
How can autoformalization preserve diverse faithful statements that improve prover search under a fixed budget?
A single formal translation can hide other faithful formulations, while syntactic differences among equivalent statements can change how a prover searches. Correctness-only, single-output evaluation therefore misses effects that matter for downstream proving.
How can BDD variable ordering minimize quantum circuit cost when BDD size is a poor proxy?
Reversible synthesis maps Boolean functions to quantum circuits, but the BDD variable ordering can substantially change the resulting circuit. An ordering that produces a small BDD may still yield an expensive quantum circuit.
How can best-effort HPC preserve reproducible digital-evolution results under hardware faults?
Best-effort execution relaxes deterministic computing to cope with constrained on-device storage, data movement, and failures across many components. Variation in execution and recorded history can create artifacts that resemble genuine evolutionary effects.
How can black-box simulators be calibrated online when observations change regimes that fitness cannot reliably detect?
Sequential observations change the calibration window and therefore the simulator’s effective objective. Fitness shifts alone cannot reliably distinguish a regime change from ordinary optimization variation or indicate how parameters should adapt.
How can black-box systems detect and mitigate reward hacking in self-evolving language-model loops?
A self-evolving loop can improve its visible score while the intended capability stagnates or deteriorates. Monitoring must therefore identify proxy exploitation and support corrective update selection without relying on internal model signals.
How can compact device surrogates stay physically consistent and accurate across unseen process or geometry splits with scarce data?
Sparse device data make it difficult to learn behavior that remains reliable when fabrication processes or device geometries differ from those observed during training. The surrogate must capture physically consistent trends while extrapolating across these distribution shifts.
How can constitutive models for multi-material 3D-printed materials capture composition- and rate-dependent behavior while preserving thermodynamic consistency?
Varying the mixture of stiff and compliant constituents changes apparent stiffness, nonlinear response, and rate-dependent dissipation. A single model must represent these coupled effects across compositions without sacrificing thermodynamic consistency.
How can continuum models compute stable, highly oscillatory smectic states in complex confinement?
Smectic states couple orientational order with periodic density layers. In complex confinement, resolving these high-frequency layers while finding stable configurations strains continuum discretizations and standard relaxation methods.
How can CPU-oriented multiobjective evolutionary algorithms be tensorized without changing their optimization behavior?
MOEAs expose population-level parallelism, but mature CPU implementations often express their states, dependencies, and updates through sequential program structures. Reorganizing these computations for tensor hardware can change the operators or update logic that define the optimization algorithm.
How can deep continuous-time recurrent networks avoid depth-induced gradient attenuation under truncated temporal backpropagation?
In deep continuous-time recurrent stacks, temporal integration can delay lower-layer signals while weakening top-down learning signals across depth. Truncating temporal backpropagation can make this optimization problem more pronounced.
Previous
1 / 5
Next