---
title: 'Slate-GLM-TS: Efficient Logistic Slate Bandits'
url: https://www.emergentmind.com/topics/slate-glm-ts
type: topic
---

# Slate-GLM-TS: Efficient Logistic Slate Bandits

Slate-GLM-TS is an efficient Thompson Sampling algorithm for the logistic contextual slate bandit problem, where sequential slate selection is performed from exponentially large combinatorial action spaces, and only a single binary reward is observed per round. It operates under a global logistic reward model, leveraging local planning with independent slot-wise maximization and global learning via joint parameter estimation, and achieves both low regret and polynomial per-round computational complexity under mild diversity assumptions [2506.13163].

## 1. Problem Formulation: Logistic Contextual Slate Bandits

The logistic contextual slate bandit framework models a sequential decision process over $T$ rounds. At each time step $t$, for each of $N$ slots, the agent faces a finite, potentially different item set $\mathcal X^i_t \subset \mathbb R^d$ and selects one item $x_t^i \in \mathcal X^i_t$. The chosen slate $x_t = (x^1_t, ..., x^N_t)$ lies in the Cartesian product $\mathcal X_t^1 \times \cdots \times \mathcal X_t^N$. A single binary reward $y_t \in \{0, 1\}$ is observed, drawn according to the probability $\mathbb P[y_t=1\mid x_t] = \mu(x_t^\top \theta^*)$ where $\mu(u) = 1/(1+e^{-u})$, and the true parameter $\theta^* \in \mathbb R^{dN}$ with $\|\theta^*\|\leq S$ is unknown.

The objective is to minimize cumulative regret over $T$ rounds:
\[
R(T) = \sum_{t=1}^T \Bigg( \max_{x \in \mathcal X_t^1\times\cdots\times\mathcal X_t^N} \mathbb E[\mu(x^\top \theta^*)] - \mathbb E[\mu(x_t^\top\theta^*)] \Bigg).
\]
The analysis assumes a slot-wise diversity condition: conditioned on previous history, $\mathbb E[x_t^i|\mathcal F_t]=0$ and $\mathbb E[x_t^i(x_t^i)^\top|\mathcal{F}_t] \succeq \rho \kappa I_d$ for each slot, with $\kappa$ dependent on the logistic curvature.

## 2. Algorithmic Structure of Slate-GLM-TS

Slate-GLM-TS adopts the Thompson Sampling (TS) paradigm, combining a Gaussian-style random perturbation of the regularized logistic-GLM estimator with local slot-wise greedy action selection. Parameter estimation relies on cumulative slot-wise statistics and is updated via an adaptive optimism-in-the-face-of-uncertainty (OFU) subroutine.

Algorithmic workflow at each round $t$:

1. **Context Observation:** Observe item sets $\{\mathcal X^i_t\}_{i=1}^N$ for all slots.
2. **Posterior Sampling:** Draw $\widetilde\theta_t = \hat\theta_t + \sigma_t U_t^{-1/2} \eta$, where $U_t = \mathrm{diag}((W^1_t)^{1/2}, ..., (W^N_t)^{1/2})$, $\sigma_t = O(\sqrt{Nd\log(t/\delta)})$, and $\eta \sim \mathcal{D}^{TS}$ satisfies TS distributional guarantees.
3. **Local Planning:** For each slot $i$, extract the $i$th block $\widetilde\theta^i_t$ and select $x_t^i = \arg\max_{x \in \mathcal X^i_t} \langle x, \widetilde\theta^i_t\rangle$. The slate $x_t$ is the tuple of these choices.
4. **Action and Update:** Execute slate $x_t$, observe reward $y_t$, and update posterior parameters via the adaptive-OFU subroutine (recomputing $\hat\theta_{t+1}$, $\{W^{i}_{t+1}\}$, $W_{t+1}$, and $\Theta_{t+1}$).

The algorithm enforces that each sampled $\widetilde\theta_t$ falls within the admissible confidence set $\Theta_t$. The following table summarizes the action selection and update steps:

| Step              | Computation                        | Purpose                      |
|-------------------|------------------------------------|------------------------------|
| Posterior sample  | $\widetilde\theta_t$ as above      | TS for exploration           |
| Slot selection    | $x_t^i = \arg\max \langle \cdot, \widetilde\theta^i_t\rangle$ | Local maximization           |
| Posterior update  | Update via adaptive-OFU            | Keep $\theta^*$ in $\Theta_t$|

## 3. Regret Bound and Theoretical Guarantees

In the fixed-arm scenario, Slate-GLM-TS-Fixed achieves (under the stated diversity assumption) $\widetilde O(\sqrt T)$ expected regret. Specifically, if the minimal eigenvalue of each slot-wise design matrix $W_t^i$ grows at least linearly in $t$ ($\lambda_{\min}(W_t^i) \gtrsim \rho t$), then employing:

- Local-global design equivalence ($W_t \approx \mathrm{diag}(W^1_t, ..., W^N_t)$),
- TS posterior concentration ($\|\widetilde\theta_t - \hat\theta_t\|_{W_t} \leq O(\sqrt{Nd\log T})$),
- Elliptical potential arguments ($\sum \|x_t\|_{W_t^{-1}}^2 = O(d\log T)$),

one can establish
\[
R(T) \leq \tilde{O}\Big(S N^{3/2} d^{3/2} \sqrt{T \log T}\Big)
\]
for fixed-arm settings, with the contextual bandit regret analysis relying on analogous design equivalence and confidence set invariance. The contextual case is conjectured to match this bound when design diversity holds.

## 4. Computational Complexity

Slate-GLM-TS ensures each round requires only polynomial time in the number of slots $N$ and feature dimension $d$. Action selection in round $t$ entails $N$ independent maximizations over $|\mathcal X^i_t|$, with total time $\sum_{i=1}^N O(|\mathcal X^i_t|) = N^{O(1)}$. Posterior updates, including recalculation of parameter estimates and confidence sets, only require $(dN)^{O(1)}\log T$ time. The exponential cost associated with evaluating all possible slates is thus replaced by scalable local maximization.

## 5. Comparison to Slate-GLM-OFU

Slate-GLM-TS and Slate-GLM-OFU both exploit the multiplicative-equivalence property of slot-wise and global slate design. However, the two algorithms differ in their approach:

- **Slate-GLM-OFU:** Maintains an explicit optimistic confidence set $\Theta_t$ and computes the maximizer of an upper confidence bound (UCB) over the slate, relying on projection into $\Theta_t$. The optimization decomposes into slot-level maximizations due to design equivalence.
- **Slate-GLM-TS:** Samples a perturbed parameter within the confidence set and performs greedy slot-wise action selection. It requires no UCB maximization, only TS sampling.
- **Regret:** OFU provides a clean high-probability $\widetilde O(dN\sqrt{T})$ guarantee for the contextual problem, while Slate-GLM-TS matches $\widetilde O(\sqrt T)$ in the fixed-arm case with conjectured analogous guarantees under diversity for the contextual scenario.
- **Implementation:** TS is simpler when a valid posterior sampling distribution $\mathcal D^{TS}$ is available and does not require explicit confidence set optimization as in the OFU approach.
- **Empirical Performance:** Both methods demonstrate low regret and competitive runtime in experiments; TS sometimes explores more naturally and is simpler to code.

## 6. Practical Applications and Experimental Evidence

Slate-GLM-TS has been empirically validated across synthetic experiments, consistently outperforming state-of-the-art baselines in both cumulative regret and runtime. One key application area is the automated selection of in-context examples for large language model prompts in binary classification tasks, such as sentiment analysis, where it achieves competitive test accuracy for the end task [2506.13163]. The algorithm’s ability to efficiently handle exponentially large action sets via slot-wise decomposition makes it viable in both research and practical deployments.

Source: https://www.emergentmind.com/topics/slate-glm-ts