Gradient-Informed Placement

• Gradient-informed placement is a method that leverages derivative signals from a governing field or surrogate model to efficiently choose placement locations.
• It applies gradient cues from interference fields, Fisher information, or learned surrogates to refine decisions in wireless networks, sensor design, and VLSI layouts.
• The approach yields faster convergence and improved performance, demonstrated by enhanced network capacity, optimized chip routing, and efficient parameter selection in ML models.

Gradient-informed placement denotes a class of placement methodologies in which decisions about where to place physical assets, sensors, graph vertices, circuit cells, computational routes, or trainable parameters are guided by gradient information associated with a governing field, an objective, a surrogate model, or an information criterion. In the surveyed literature, the guiding gradient may be the spatial gradient of an interference field, the derivative of a confidence-region criterion with respect to a design measure, the derivative of a contact-wrench residual with respect to geometry, the gradient of a learned surrogate constrained by prior beliefs, or gradients with respect to discrete one-hot inputs and masked adapter entries (Malik et al., 2011, Neitzel et al., 2019, Boige et al., 2023, Aglietti et al., 2024, Li et al., 2024, Liu et al., 24 Feb 2025, Sehanobish et al., 12 May 2026).

1. Scope and formal structure

Across domains, gradient-informed placement couples a placement variable with a differentiable signal that ranks local moves. The placement variable may be a continuous point in space, a geometric parameter vector, a positive Borel measure over an experimental domain, a discrete genotype represented in one-hot form, a routing decision in a sparse network, or a binary mask over trainable adapter entries. The central operation is to use derivatives to identify directions, locations, or subsets that are expected to improve a placement objective more efficiently than uninformed search.

A common mathematical pattern is an optimization problem of the form “choose placements so as to minimize or maximize a scalar criterion,” then replace direct search by local or surrogate gradient information. In wireless deployment this criterion is reduced to minimizing an interference field $g(\mathbf{z})$ (Malik et al., 2011). In PDE-constrained sensor design it is a convex functional of a Fisher-information operator $\mathcal{I}(\omega)$ over a design measure $\omega$ (Neitzel et al., 2019). In graph drawing it is the Fruchterman–Reingold energy or its attractive-only surrogate, combined with coordinate Newton directions (Hamaguchi et al., 2024). In discrete illumination it is a stochastic scalarization $g(x)=|w_0|f(x)+\sum_i w_i c_i(x)$, differentiated with respect to one-hot inputs (Boige et al., 2023). In LoRA parameter placement it is a score derived from initialization gradients, such as $\hat S_{ij}=\left|\frac1N\sum_k g_{k,ij}\right|$ or $\hat F_{ij}=\frac1N\sum_k g_{k,ij}^2$ (Sehanobish et al., 12 May 2026).

Domain	Placement target	Gradient signal
Wireless downlink	Added base-station locations	$\nabla g(\mathbf{z})$ of interference
PDE-constrained sensing	Sensor measure $\omega$	$\psi'(\omega)(x)$ from Fisher information
Contact-rich robotics	Geometry estimate used for placement	$\partial c_t/\partial \theta$
Electromagnetic design	Geometric design parameters $\mathcal{I}(\omega)$0	$\mathcal{I}(\omega)$1, $\mathcal{I}(\omega)$2
Graph and chip layout	Coordinates of vertices or cells	Wirelength/density gradients, coordinate Newton directions
Discrete QD	Offspring placement in archive cells	$\mathcal{I}(\omega)$3 over one-hot inputs
MoE and LoRA	Expert routes or trainable entries	Routing gradients; $\mathcal{I}(\omega)$4, $\mathcal{I}(\omega)$5

This breadth suggests that “placement” is not restricted to Euclidean coordinates. A plausible implication is that the term is best understood operationally: gradient-informed placement is any placement procedure in which the decisive local signal comes from derivatives rather than from exhaustive combinatorial search or purely random heuristics.

2. Field-driven placement in physical space

In "Optimal Base Station Placement: A Stochastic Method Using Interference Gradient In Downlink Case" (Malik et al., 2011), the placement variable is the location of additional base stations in a downlink network. The key field is the interference function

$\mathcal{I}(\omega)$6

with gradient

$\mathcal{I}(\omega)$7

The method partitions the region of interest by Delaunay triangulation and proves that, inside each triangle, $\mathcal{I}(\omega)$8 is convex away from the vertices; therefore any interior local minimum is the unique global minimum in that triangle. Gradient descent is initialized at the triangle centroid, projected to remain inside the triangle, and yields one candidate minimum-interference point per triangle. Two selection strategies are then considered: a one-shot ranking of all candidates, and a sequential scheme that recomputes the triangulation and minima after each added base station. In the reported $\mathcal{I}(\omega)$9 scenario, the second heuristic improves average capacity from $\omega$0 to $\omega$1 bits/s/Hz/km$\omega$2 and coverage from $\omega$3 to $\omega$4 (Malik et al., 2011).

In "A sparse control approach to optimal sensor placement in PDE-constrained parameter estimation problems" (Neitzel et al., 2019), placement is formulated over measures rather than points. A sensor design is a positive Borel measure

$\omega$5

and the Fisher information is

$\omega$6

The reduced design functional is $\omega$7, and its derivative is the continuous field

$\omega$8

The conditional-gradient method places new sensing mass at locations minimizing $\omega$9, i.e. where marginal information gain is maximal. The measure formulation is sparse in a precise sense: there exists an optimal solution with support size at most $g(x)=|w_0|f(x)+\sum_i w_i c_i(x)$0 (Neitzel et al., 2019). Here, gradient-informed placement is not a geometric descent in $g(x)=|w_0|f(x)+\sum_i w_i c_i(x)$1 but an active-set construction in measure space.

In "Stable Object Placement Under Geometric Uncertainty via Differentiable Contact Dynamics" (Li et al., 2024), stable placement is achieved by continuously updating uncertain geometry through gradients of a force–torque residual. A differentiable simulator provides

$g(x)=|w_0|f(x)+\sum_i w_i c_i(x)$2

and the geometry update minimizes

$g(x)=|w_0|f(x)+\sum_i w_i c_i(x)$3

by gradient descent on $g(x)=|w_0|f(x)+\sum_i w_i c_i(x)$4. Because contact is hybrid and sensitive to initialization, the method maintains a belief $g(x)=|w_0|f(x)+\sum_i w_i c_i(x)$5 over multiple hypotheses and selects the lowest-cost geometry estimate for control. The reported tasks include in-hand pose uncertainty, wall-height uncertainty, pillar location and height uncertainty, and coffee-cup placement on a saucer (Li et al., 2024). The placed object is controlled by a policy $g(x)=|w_0|f(x)+\sum_i w_i c_i(x)$6, but the placement becomes gradient-informed because the geometry conditioning that defines the target stable set is itself inferred through differentiable contact dynamics.

3. Surrogate-based placement and parametric design

A second major interpretation uses gradients not to move the object directly, but to shape a surrogate on which placement decisions are made. "GradINN: Gradient Informed Neural Network" (Aglietti et al., 2024) introduces two coupled networks: a primary surrogate $g(x)=|w_0|f(x)+\sum_i w_i c_i(x)$7 for the unknown field $g(x)=|w_0|f(x)+\sum_i w_i c_i(x)$8, and an auxiliary network $g(x)=|w_0|f(x)+\sum_i w_i c_i(x)$9 that expresses prior beliefs about $\hat S_{ij}=\left|\frac1N\sum_k g_{k,ij}\right|$0. For scalar output, the loss is

$\hat S_{ij}=\left|\frac1N\sum_k g_{k,ij}\right|$1

with

$\hat S_{ij}=\left|\frac1N\sum_k g_{k,ij}\right|$2

The framework is explicitly proposed for low-data regimes in which prior beliefs about smoothness or gradient regularity are easier to specify than governing equations. The paper is not about placement in the narrow spatial sense, but it directly frames placement as a downstream use case: once a surrogate with controlled gradients is learned, one can optimize placements, arrangements, or design variables using $\hat S_{ij}=\left|\frac1N\sum_k g_{k,ij}\right|$3 rather than expensive simulations (Aglietti et al., 2024).

"Gradient-Informed Machine Learning in Electromagnetics" (Zorzetto et al., 26 Jan 2026) instantiates this strategy for a permanent magnet synchronous machine parameterized by $\hat S_{ij}=\left|\frac1N\sum_k g_{k,ij}\right|$4. Isogeometric Analysis provides both the field solution $\hat S_{ij}=\left|\frac1N\sum_k g_{k,ij}\right|$5 and parametric sensitivities $\hat S_{ij}=\left|\frac1N\sum_k g_{k,ij}\right|$6, which are then used in Proper Orthogonal Decomposition and Gaussian Process Regression. The reduced coefficients satisfy

$\hat S_{ij}=\left|\frac1N\sum_k g_{k,ij}\right|$7

and gradient-enhanced GPR is trained either on these coefficients or directly on torque $\hat S_{ij}=\left|\frac1N\sum_k g_{k,ij}\right|$8 and its derivatives. The full IGA model has $\hat S_{ij}=\left|\frac1N\sum_k g_{k,ij}\right|$9 DOFs, each simulation takes about $\hat F_{ij}=\frac1N\sum_k g_{k,ij}^2$0 s, and direct torque GPR training ranges from $\hat F_{ij}=\frac1N\sum_k g_{k,ij}^2$1 to $\hat F_{ij}=\frac1N\sum_k g_{k,ij}^2$2 s while field POD-GPR training ranges from $\hat F_{ij}=\frac1N\sum_k g_{k,ij}^2$3 to $\hat F_{ij}=\frac1N\sum_k g_{k,ij}^2$4 s depending on sample count (Zorzetto et al., 26 Jan 2026). The paper’s principal claim is that parametric sensitivities materially improve sample efficiency; a plausible implication is that this kind of surrogate is especially well suited to geometric placement problems in which repeated gradient evaluations are required.

4. Layout, routing, and coordinate placement on graphs and chips

Several works treat placement itself as the primary optimization variable and use gradients of explicit layout objectives. "FFTPL: An Analytic Placement Algorithm Using Fast Fourier Transform for Density Equalization" (Lu et al., 2013) formulates global VLSI placement as minimization of smoothed wirelength plus an electrostatic density energy: $\hat F_{ij}=\frac1N\sum_k g_{k,ij}^2$5 Density is modeled by Poisson’s equation

$\hat F_{ij}=\frac1N\sum_k g_{k,ij}^2$6

with Neumann boundary conditions, and the electric field $\hat F_{ij}=\frac1N\sum_k g_{k,ij}^2$7 supplies the density gradient. The resulting nonlinear placement is solved by conjugate gradient with FFT-based Poisson solves of complexity $\hat F_{ij}=\frac1N\sum_k g_{k,ij}^2$8. On the ISPD 2005 suite, FFTPL improves average total wirelength by $\hat F_{ij}=\frac1N\sum_k g_{k,ij}^2$9 over Capo10.5, $\nabla g(\mathbf{z})$0 over FastPlace3.0, $\nabla g(\mathbf{z})$1 over APlace2, $\nabla g(\mathbf{z})$2 over NTUPlace3, and $\nabla g(\mathbf{z})$3 over mPL6 (Lu et al., 2013).

"The Power of Graph Signal Processing for Chip Placement Acceleration" (Liu et al., 24 Feb 2025) recasts placement as smoothness minimization on a circuit graph. With graph Laplacian $\nabla g(\mathbf{z})$4, the quadratic wirelength surrogate is the Laplacian smoothness

$\nabla g(\mathbf{z})$5

GiFt approximates the ideal low-pass filter $\nabla g(\mathbf{z})$6 by a parameter-free multi-resolution graph filter,

$\nabla g(\mathbf{z})$7

and uses the filtered signal as an initialization for analytical placement. Relative to DREAMPlace, the reported effects are essentially unchanged HPWL, a $\nabla g(\mathbf{z})$8 reduction in iterations, and about $\nabla g(\mathbf{z})$9 runtime reduction (Liu et al., 24 Feb 2025). Although GiFt does not backpropagate through a learned model, it is gradient-informed in the sense that it analytically preconditions the directions that subsequent gradient-based placers would otherwise need to discover iteratively.

"Placement Optimization with Deep Reinforcement Learning" (Goldie et al., 2020) moves the gradients from placement coordinates to policy parameters. The policy objective for a graph $\omega$0 is

$\omega$1

with policy gradient

$\omega$2

Here, placement is a sequence of graph-to-location assignments, and gradients are taken with respect to the neural policy rather than through a differentiable placement cost (Goldie et al., 2020). The surveyed text emphasizes that this allows optimization of non-differentiable rewards such as runtime or routed congestion.

"Initial Placement for Fruchterman--Reingold Force Model With Coordinate Newton Direction" (Hamaguchi et al., 2024) develops a gradient- and Hessian-based initial placement for graph drawing. The Fruchterman–Reingold energy is optimized first in a discrete attractive-only surrogate on a hexagonal lattice, then refined by FR or L-BFGS. For a selected vertex $\omega$3, the update is

$\omega$4

where the Newton direction is computed from the gradient and Hessian of the attractive objective. The method is motivated by the observation that direct coordinate Newton on the full FR objective is unreliable, whereas the attractive-only per-vertex objective is strictly convex (Hamaguchi et al., 2024).

5. Discrete placement inside search spaces and models

Gradient-informed placement is equally prominent in intrinsically discrete settings. "Gradient-Informed Quality Diversity for the Illumination of Discrete Spaces" (Boige et al., 2023) treats the placement of offspring into MAP-Elites archive cells as a gradient-guided process. A stochastic scalarization

$\omega$5

is differentiated with respect to a one-hot representation of the discrete genotype. For a one-symbol mutation $\omega$6, the local improvement is approximated by

$\omega$7

and a Boltzmann distribution over discrete neighbors is formed from $\omega$8. The reported average Pearson correlation between $\omega$9 and true discrete improvements in Discrete LSI is $\psi'(\omega)(x)$0, and ME-GIDE outperforms random mutation and projected continuous baselines across protein design and discrete latent-space illumination benchmarks (Boige et al., 2023). The key point is that archive placement in descriptor space is no longer a random walk; it is biased by gradients of both fitness and descriptors.

"GRIN: GRadient-INformed MoE" (Liu et al., 2024) applies the same principle to the placement of tokens among experts in a Mixture-of-Experts model. Instead of treating deterministic top-$\psi'(\omega)(x)$1 routing as non-differentiable, GRIN uses SparseMixer-v2, a straight-through-style sparse gradient estimator for routing. This makes the routing decision itself directly informed by gradients of the language-model loss. The reported top-2 $\psi'(\omega)(x)$2B MoE has $\psi'(\omega)(x)$3B total parameters and $\psi'(\omega)(x)$4B activated parameters, and achieves $\psi'(\omega)(x)$5 on MMLU, $\psi'(\omega)(x)$6 on HellaSwag, $\psi'(\omega)(x)$7 on HumanEval, and $\psi'(\omega)(x)$8 on MATH (Liu et al., 2024). The same work also ties routing quality to systems design: model parallelism is configured to avoid token dropping, so the learned routing decisions are not invalidated by capacity-factor heuristics.

"Not How Many, But Which: Parameter Placement in Low-Rank Adaptation" (Sehanobish et al., 12 May 2026) addresses placement inside a LoRA adapter. With $\psi'(\omega)(x)$9 and frozen $\partial c_t/\partial \theta$0, each candidate trainable entry $\partial c_t/\partial \theta$1 is scored from initialization gradients using either

$\partial c_t/\partial \theta$2

Under supervised fine-tuning, random and informed subsets are comparable because gradients are described as low-rank and directionally stable. Under GRPO on base models, random placement fails to improve over the base model, while gradient-informed placement recovers standard LoRA accuracy. On Qwen2.5-1.5B for GSM8K, the base model scores $\partial c_t/\partial \theta$3, random $\partial c_t/\partial \theta$4K gives $\partial c_t/\partial \theta$5, while $\partial c_t/\partial \theta$6 and $\partial c_t/\partial \theta$7 circuits reach $\partial c_t/\partial \theta$8 and $\partial c_t/\partial \theta$9, compared to $\mathcal{I}(\omega)$00 for full LoRA (Sehanobish et al., 12 May 2026). Selected parameters concentrate on V, O, and Down projections, described in the paper as residual-stream-writing projections.

A software-level variant appears in "JaxDecompiler: Redefining Gradient-Informed Software Design" (Pochelu, 2024). There, placement refers to the organization of gradient-generated computations across modules, platforms, and communication layers. By decompiling JAX gradient functions into editable Python, the work makes it possible to alter the placement of guards, all-reduce operations, or parallel primitives based on observed gradient behavior. This suggests that gradient-informed placement can also mean placement of computation, not only placement of objects.

6. Shared algorithmic themes, empirical effects, and limits

Several algorithmic motifs recur. One is objective re-expression: base-station placement replaces expected-capacity maximization by interference minimization (Malik et al., 2011), GiFt replaces repeated nonlinear placement iterations by a graph filter approximating $\mathcal{I}(\omega)$01 (Liu et al., 24 Feb 2025), and ME-GIDE replaces exhaustive neighbor evaluation by first-order Taylor scores (Boige et al., 2023). Another is local convexification: Delaunay triangles isolate unique interference minima (Malik et al., 2011), the attractive-only per-vertex graph-drawing objective is strictly convex (Hamaguchi et al., 2024), and the sensor-design problem becomes convex in measure space (Neitzel et al., 2019). A third is iterative refinement under changing geometry or support: Heuristic 2 in the wireless setting recomputes minima after each addition (Malik et al., 2011), the robotics method updates a belief over geometries at every control step (Li et al., 2024), and sparse conditional-gradient methods add one support point at a time in sensor placement (Neitzel et al., 2019).

The empirical effects are domain-specific but structurally similar: fewer evaluations of expensive objectives, faster convergence to useful configurations, or strong performance at very small effective budgets. Representative results include the coverage increase from $\mathcal{I}(\omega)$02 to $\mathcal{I}(\omega)$03 in downlink BS placement (Malik et al., 2011), GiFt’s runtime reduction of about $\mathcal{I}(\omega)$04 relative to DREAMPlace with essentially unchanged HPWL (Liu et al., 24 Feb 2025), and the recovery of full-LoRA-level GRPO performance with sparse gradient-selected circuits (Sehanobish et al., 12 May 2026). In MoE training, gradient-informed routing plus no token dropping enables a $\mathcal{I}(\omega)$05B-active-parameter model to match a $\mathcal{I}(\omega)$06B dense model trained on the same data (Liu et al., 2024).

Several misconceptions are not supported by the surveyed works. Gradient-informed placement does not necessarily mean direct differentiation of the final objective with respect to Euclidean coordinates. In RL placement, gradients are taken with respect to policy parameters (Goldie et al., 2020). In GradINN, gradients constrain a surrogate rather than a placement variable itself (Aglietti et al., 2024). In sensor design, the optimization variable is a measure, and the gradient is a scalar field over admissible locations (Neitzel et al., 2019). Nor does gradient information guarantee robustness: GradINN oversmooths steep gradients in Burgers’ equation under a smooth prior (Aglietti et al., 2024), robotic geometry estimation remains sensitive to initialization and therefore uses multiple hypotheses (Li et al., 2024), and GRPO parameter placement is sharply regime-dependent because high-rank near-orthogonal gradients make random masks ineffective (Sehanobish et al., 12 May 2026).

Taken together, these works suggest a unifying interpretation. Gradient-informed placement is not a single algorithmic family but a design principle: expose a placement-relevant differential signal, use that signal to reduce an otherwise hard search space, and couple the resulting local information to a sparse, sequential, or surrogate-based optimizer. The principle appears in continuous geometry, discrete combinatorial spaces, control policies, scientific machine learning, and internal neural architectures, and its effectiveness depends less on the nominal domain than on whether the chosen gradient faithfully captures the structure of the placement objective.