---
title: One-by-One Sequential Processing
url: https://www.emergentmind.com/topics/one-by-one
type: topic
---

# One-by-One Sequential Processing

“One-by-One” denotes a recurring research pattern in which individual units are processed sequentially, locally, or with explicit per-item state, rather than only through aggregate operators. Across the cited literature, the expression is used for diagonal input gating in nonlinear feature selection, progressive collection of weakly supervised action evidence, ordered enumeration of tagged visual objects, sequential screening and no-recall bandit decisions, single-particle partitioning and absorption, one-at-a-time extraction of atoms, per-defect and per-pulsar diagnostics, and day-by-day assignment through balanced permutation sequences [2108.12165] [1807.02929] [2404.16375] [2301.13397] [1202.6243] [1611.09628] [2411.14846] [2602.21687].

## 1. Semantic scope and recurring formalizations

In the cited work, “one-by-one” is not a single technical term but a family of constructions. Sometimes it means literal sequential processing of one entity at a time, as in vehicular edge access where only one vehicle wakes up per frame, or in online bandits where a current arm can be pulled or permanently abandoned [2301.13460] [1707.04987]. In other settings it denotes a structural decomposition into independent units, such as one-to-one input gates in LassoLayer or one-by-one clause imposition in compressed ALLSAT [2108.12165] [1608.08472]. In physical sciences it often refers to elementary excitations or particles being partitioned, absorbed, or extracted individually [1202.6243] [2010.02676] [2002.01767].

| Domain | Operational meaning | Representative paper |
|---|---|---|
| Feature selection | One input dimension, one gate | [2108.12165] |
| Weak supervision | One classifier stage collected after another | [1807.02929] |
| Multimodal LLMs | List tagged objects in numeric order | [2404.16375] |
| Sequential decision | One test, arm, or vehicle at a time | [2301.13397], [1707.04987], [2301.13460] |
| Quantum and condensed matter | One excitation, particle, atom, or trap event at a time | [1202.6243], [2010.02676], [2002.01767], [1010.4456] |
| Diagnostics and fairness | One defect, pulsar, or daily assignment step at a time | [1611.09628], [2411.14846], [2602.21687] |

A closely related formulation is “one-to-one.” In LassoLayer, the first layer is a diagonal operator with one scalar gate per input coordinate, while in asynchronous network theory there is a one-to-one correspondence between effective connectivity and the temporal structure of pairwise averaged correlations [2108.12165] [1411.4770]. The common feature is explicit locality: each unit has its own gate, test, trajectory, clause state, or assignment position.

## 2. Learning systems: gating, ordered supervision, and target-conditioned explanations

In nonlinear feature selection, LassoLayer turns selection into a learned diagonal gating operation. For input $x \in \mathbb{R}^d$, the layer applies
$$
y=\sigma_{\mathrm{out}}\!\big(w\odot \sigma_{\mathrm{in}}(x)\big),
$$
and in the experimental LassoMLP instantiation both activations are identity, so the layer reduces to $y=w\odot x$ [2108.12165]. Sparsity is imposed only on the gate vector $w$ through
$$
\mathcal{L}(\theta,w)=\mathcal{L}_{\mathrm{task}}(\theta,w)+\lambda\lVert w\rVert_1,
$$
followed by the proximal shrinkage operator
$$
S_\lambda(w_i)=\operatorname{sign}(w_i)\max(|w_i|-\lambda,0),
$$
which yields exact zeros and therefore true “off” switches [2108.12165]. The paper also introduces an early-phase “kicking” heuristic: during the first $K$ epochs, any zero gate is reactivated with probability $\rho$ by setting $w_i:=\pm\delta$. In the reported settings, $K=1000$, $\rho=0.1$, and $\delta=0.1$ for synthetic data or $\delta=0.25$ for MNIST [2108.12165]. On a nonlinear regression task with $x\in\mathbb{R}^{256}$ and output $y=\sin(x_1)\exp(-x_2)+(x_3-0.2)^2$, LassoMLP’s feature-selection AUC approaches $\approx 1.0$ as $N$ increases, whereas linear Lasso saturates near $\approx 0.8$ for $N\ge 100$ [2108.12165]. On MNIST with 784 image dimensions augmented by 5000 Gaussian noise features, LassoMLP outperforms LassoNet, HSIC Lasso, PFA, and Fisher score in all but two reported settings, namely $(210,40)$ and $(3500,10)$ for training samples and selected features [2108.12165].

A different learning use of one-by-one supervision appears in Set-of-Mark training for multimodal LLMs. “List items one by one” asks the model to enumerate every tagged object in ascending numeric order, producing outputs such as “1. person, 2. cat, 3. dog” [2404.16375]. The task is trained with standard autoregressive next-token prediction rather than explicit alignment or ordering losses. The dataset construction reported in the paper uses 10k SoM-listing pairs and 20k SoM-QA conversations on MS-COCO, mixed with LLaVA-1.5 Mix665K for a total of 695k instruction-tuning samples [2404.16375]. The reported gains persist even when tags are omitted at inference: for LLaVA-1.5 Vicuna-13B, POPE F1 improves from 85.9 to 86.6, MME overall from 1531.3 to 1563.1, SEED-I from 68.2 to 69.6, and LLaVA-W from 70.7 to 75.3 [2404.16375]. With tags at test time, MM-Vet reaches 37.2, which is +1.8 over the baseline LLaVA-1.5 13B [2404.16375].

In explanation methods, GANMEX makes one-vs-one attributions depend on a generated target-class counterfactual baseline rather than on a zero or blur baseline. The generator $G(x,c)$ is trained so that the baseline is realistic, classified as the target class, and close to the original input; the classifier being explained is inserted into the adversarial objective as the class discriminator [2011.06015]. For Integrated Gradients, the attribution is computed on the score difference $S_{c_o}(z)-S_{c_t}(z)$ along the path from the GANMEX baseline $\tilde x$ to $x$. Reported quantitative gains include AOPC100 on MNIST improving from 0.614 with a zero baseline to 1.421 with GANMEX, and inverse localization on BAM improving overall from 1.591 with zero and 0.852 with MDTS to 0.747 with GANMEX [2011.06015]. The paper also reports better sensitivity under cascading randomization of the classifier, addressing a standard criticism of saliency methods [2011.06015].

## 3. Sequential discovery, screening, and resource allocation

In weakly supervised temporal action detection, “one-by-one” names the testing phase of a two-stage method: step-by-step erasion during training and one-by-one collection during inference [1807.02929]. A snippet-wise classifier is trained, the most discriminative clips are stochastically erased, a new classifier is trained on the remainder, and the process repeats up to four steps [1807.02929]. At test time, all trained classifiers are run, their soft masks and probabilities are fused, and a fully connected CRF refines temporal continuity [1807.02929]. On THUMOS’14, the reported weakly supervised detector reaches mAP@0.5 of 15.9 versus 13.7 for UntrimmedNet; on ActivityNet v1.2 it achieves average mAP 15.6, with mAP@0.5 of 27.3 [1807.02929]. The FC-CRF is a major part of the one-by-one collection stage: on THUMOS’14 it raises mAP@0.5 from 6.9% to 14.0%, and on ActivityNet it raises average mAP from 2.6% to 14.9% [1807.02929].

In sequential strategic screening, one-by-one means a fixed pipeline of classifiers through which an agent can manipulate between stages [2301.13397]. The central result is the “zig-zag” phenomenon: because earlier-stage constraints need not be maintained later, sequential success can be strictly cheaper than satisfying the intersection of all positive regions simultaneously [2301.13397]. In the paper’s two-dimensional example, the conjunctive best response costs $\sqrt{2}$, while the sequential best response costs $5/4$ by first moving to $(0,1/4)$ and then to $(1,1/4)$ [2301.13397]. More generally, the paper proves $\operatorname{conj}(x,\{h_1,\dots,h_k\})\ge \operatorname{seq}(x,\{h_1,\dots,h_k\})$, gives a convex program for optimal sequential manipulation, shows an unbounded gap in two dimensions, and proves that monotone classifiers eliminate zig-zag advantages [2301.13397].

In the online streaming multi-armed bandit model, arms arrive one at a time and cannot be revisited after being skipped [1707.04987]. For Bernoulli arms with means drawn i.i.d. from a distribution with left-tail behavior $F(x)=C x^m+o(x^m)$ near zero, the paper derives lower bounds of order $N^{-1/m}$ in the large-$K$ regime and $K^{-1/(m+1)}$ in the small-$K$ regime, and matches them up to constant factors with threshold-style policies [1707.04987]. In the uniform case $F(x)=x$, the fixed-payout recursion is $b_i=b_{i-1}-(1/2)b_{i-1}^2$, giving $b_i\asymp 2/i$, while the small-$K$ lower bound is $\Omega(K^{-1/2})$ [1707.04987]. The no-revisit constraint is the defining difference from classical MAB formulations [1707.04987].

Vehicular edge computing uses the phrase more literally. The reported system adopts a one-by-one scheduling mechanism in which only one vehicle is active in uplink and only one in downlink per frame, with the optimization jointly choosing scheduling, offloading ratio, and bit allocation over a mission horizon [2301.13460]. The uplink energy for vehicle $k$ in frame $n$ is
$$
E_{k,n}^{\mathrm{one}}=a_k^u[n]\frac{N_0B\Delta}{\lVert h_k[n]\rVert^2}\left(2^{l_k^u[n]/(B\Delta)}-1\right),
$$
and the paper solves the resulting mixed-integer non-convex problem through a Lagrange dual method [2301.13460]. In the reported numerical experiments, optimized one-by-one access yields lower total vehicle energy than local execution, orthogonal access, and one-by-one access with equal bit allocation [2301.13460].

## 4. Physical one-by-one processes: partitioning, absorption, extraction, and activation

In electron quantum optics, “partitioning electrons one by one” refers to a Hanbury Brown–Twiss geometry fed by an on-demand mesoscopic capacitor that emits one electron and one hole per cycle into a quantum Hall edge channel [1202.6243]. A quantum point contact with transmission $T$ and reflection $1-T$ partitions the excitations, and the low-frequency current correlations count emitted electron/hole excitations at the single-charge level [1202.6243]. In the equal-temperature excess-noise form used experimentally,
$$
S_{34}=-\delta S_{44}=-4e^2 f_d T(1-T)\,\delta N_{\mathrm{HBT}},
$$
so at $T=1/2$ the measured $\delta S_{44}/(e^2 f_d)$ directly yields $\delta N_{\mathrm{HBT}}$ [1202.6243]. The paper also shows that thermal antibunching suppresses low-energy source excitations, making the noise a probe of the emitted energy distribution [1202.6243].

In unbound quantum dynamics, one-by-one absorption is implemented with a one-body complex absorbing potential and a Lindblad hierarchy over particle-number sectors [2010.02676]. Each absorption event is represented as a quantum jump generated by
$$
L(x)=\sqrt{2\gamma(x)}\,\hat\psi(x),
$$
and the absorbed flux is projected onto single-particle scattering states through
$$
J_\epsilon=\sqrt{2}\int dx\,\sqrt{\gamma(x)}\,\chi_\epsilon^*(x)\hat\psi(x)
$$
to obtain singly differential spectra without constructing full many-particle scattering states [2010.02676]. The explicit two-particle formulas give $\partial P_2/\partial\epsilon$ from the time-integrated reduced one-body kernel $\Phi(x,x')$ and $\partial P_1/\partial\epsilon$ from the one-particle density matrix $\rho_1(x,x';t)$ [2010.02676]. This framework is presented as naturally extensible to any number of particles, while the paper works out compact formulas for the two-particle case [2010.02676].

The attractive Bose–Hubbard model provides a different one-by-one mechanism: by weakly modulating the tunnelling rate, a single atom can be extracted from a small matter-wave soliton while the remaining atoms stay localized [2002.01767]. The drive is written as $J(t)=J_0+\delta J\cos(\omega t)$, and in the strong-coupling regime the resonance for ejecting one atom is $\omega=(N-1)|U|/\hbar$ [2002.01767]. For $N=3$, $M=149$, and $U/J=-10$, the first excited band corresponding to one free atom becomes almost fully populated within a few tens of tunnelling times, and the transition rate scales as $\Gamma\propto (\epsilon J)^2$ with fitted slope $2.07\pm 0.04$ [2002.01767]. The paper interprets the protection of the remaining cluster through energy gaps plus translation and parity-based selection rules [2002.01767].

One-by-one measurement also appears in two-qubit tomography. The reported protocol measures all density-matrix elements directly, rather than inferring off-diagonals from an ill-conditioned linear system [1506.01317]. The key theoretical property is the condition number $\kappa(A)=1$ for the reconstruction matrix, whereas the paper reports $\kappa_{\text{Pauli}}=\sqrt{2}$, $\kappa_{\text{standard 36}}=3$, $\kappa_{\text{MUB}}=\sqrt{5}$, and $\kappa_{\text{JKMW}}\approx\sqrt{60.1}$ for comparison schemes [1506.01317]. Seventeen two-qubit polarization states were reconstructed, and the optimal protocol produced the smallest uncertainty circles in trace distance and disturbance [1506.01317].

A nanoscale condensed-matter use is one-by-one trap activation in silicon nanowire transistors. As gate voltage increases, oxide traps do not generally turn on simultaneously; instead, Coulomb repulsion between occupied traps shifts neighboring trap energies and causes sequential activation [1010.4456]. The paper reports non-overlapping occupancy peaks $g_1(V_g), g_2(V_g), g_3(V_g),\dots$ for most traps, a repulsion energy $W_{45}\approx 110$ meV for a strongly coupled pair, and a noise reduction by more than one order of magnitude relative to a naive $1/f$ superposition model [1010.4456]. The effect weakens with increased channel electron density because screening reduces both the effective trap charge and inter-trap repulsion [1010.4456].

## 5. Individualized tracking, diagnostics, and fair repetition

In cosmological defect simulations, measuring monopole velocities “one by one” means identifying every monopole and antimonopole on the lattice by its topological charge and reconstructing worldlines across analysis times [1611.09628]. The method replaced earlier field-averaged estimators and resolved a long-standing ambiguity in the velocity-dependent one-scale model: the reported average velocities are $v_r=0.76\pm 0.07$ in the radiation era and $v_m=0.65\pm 0.08$ in the matter era, with no evidence for a luminal branch [1611.09628]. The same simulations also calibrate VOS parameters, with $c_r\approx 2.6$, $k_r\approx 0.9$, $c_m\approx 2.5$, and $k_m\approx 1.6$ in the corrected tables [1611.09628].

A related diagnostic strategy appears in pulsar timing arrays. NANOGrav’s 15-year analysis removes pulsars one by one and recomputes the noise-marginalized optimal statistic to test the internal consistency of a spatially correlated stochastic signal [2411.14846]. For the full 67-pulsar array, the paper reports SCNMOS Hellings–Downs $\hat A^2=6.87\times 10^{-30}$ with S/N $=4.55$, and MCNMOS Hellings–Downs $\hat A^2=5.06\times 10^{-30}$ with S/N $=2.91$ [2411.14846]. The least-noisy-first removal order exhibits abrupt changes, including a large initial drop when removing PSR J1909−3744 and later jumps when removing B1855+09 and J2317+1439, but comparison with 100 simulated arrays shows no inconsistency with a stochastic gravitational-wave background [2411.14846].

In fair division, one-by-one refers to day-by-day assignment rather than end-of-horizon averaging. The model considers $n$ players and $n$ indivisible items per day, with each day represented by a permutation of item ranks to players [2602.21687]. Strong balance requires
$$
Z_i^t[j]\le \frac{jn}{t}
$$
for every player $i$, day $t$, and rank position $j$, while weak balance relaxes this to $Z_i^t[j]\le \frac{jn}{t}+1$ [2602.21687]. The strong condition implies ordinal PROP1 after every day, exists for all $n\le 11$, and is impossible for many larger values, including all $n>61$ [2602.21687]. Weak balance also implies ordinal PROP1, exists for all $n\le 12$, and fails for $n=6k$ with $k\ge 19$ [2602.21687]. A weaker PROP2-sufficient condition remains open for general $n$ [2602.21687].

These examples show that one-by-one methods are not restricted to sparse learning or sequential control. They also function as diagnostic regimes in which the fundamental object of interest is an individual trajectory, source, or day-prefix rather than an ensemble average.

## 6. Algorithmic and theoretical consequences

Some one-by-one constructions formalize limits rather than procedures. In asynchronous recurrent networks, there is a one-to-one correspondence between effective population-level connectivity and the temporal structure of pairwise averaged correlations, except in degenerate cases [1411.4770]. For binary and LIF networks, the correlation matrices take resolvent forms such as
$$
\bar C(\omega)=(I-M(\omega))^{-1}A(I-M^T(-\omega))^{-1},
$$
and the paper proves that preserving pairwise averaged correlations under downscaling requires preserving the effective connectivity $W$ itself [1411.4770]. This yields a hard reducibility bound: when in-degree is scaled by $\kappa$, the minimal feasible factor is
$$
\kappa_{\min}=\frac{\sigma_{\mathrm{int}}^2}{\sigma_{\mathrm{int}}^2+\sigma_{\mathrm{ext}}^2},
$$
so asynchronous networks cannot in general be reduced arbitrarily while keeping both mean activity and second-order structure fixed [1411.4770].

In combinatorics and graph embedding, one-by-one appears as a constructive locality condition. A particular sub-quadtree $PQT_n$ with $(3^n-1)/2$ vertices is embedded into the $m$-dimensional crossed cube $CQ_m$ by placing vertices and edges one by one so that every quadtree edge maps to a path of length at most two [2208.11172]. The reported sufficient host dimensions are
$$
m=\left\lceil \log_2\frac{3^n-1}{2}\right\rceil \quad \text{for } n\le 8,
\qquad
m=2n-4 \quad \text{for } n>8,
$$
and the resulting embedding has dilation two [2208.11172].

Boolean satisfiability offers another clause-by-clause use. “ALLSAT compressed with wildcards” imposes clauses one by one on disjoint 012$n$-rows, using the classical don’t-care symbol 2 together with the $n$-wildcard meaning “at least one 0 here” [1608.08472]. The method maintains disjointness, so the final row family is an orthogonal DNF after optional refinement of $n$-bubbles to ordinary 012-rows [1608.08472]. The paper’s Master Theorem yields total time $O(Rh(d+s))$, and for CNFs on $w$ variables and $h$ clauses this becomes
$$
O\big((R+1)hw(h+\mathrm{sat}(F))\big),
$$
with the method reported as most efficient for few but large clauses and extendable from clauses to superclauses [1608.08472].

Across these works, one-by-one procedures usually trade aggregate symmetry for explicit local state. This suggests a common pattern: the gain is finer control—exact zeros, target-conditioned explanations, trajectory-level diagnostics, day-prefix fairness, or dilation bounds—while the cost is sensitivity to ordering, feasibility checks, or impossibility thresholds. The cited literature makes that trade-off explicit through examples such as LassoLayer’s dependence on $\lambda$ and kicking parameters, over-erasion in weakly supervised action detection, the irreversibility of no-recall bandits and sequential screening, the reducibility limit $\kappa_{\min}$ in asynchronous networks, and the nonexistence of strong balance for all sufficiently large $n$ [2108.12165] [1807.02929] [1707.04987] [2301.13397] [1411.4770] [2602.21687].

Source: https://www.emergentmind.com/topics/one-by-one