21 problems

Grid-averaging conjecture. It may be possible to relax the relevant sample-size condition by first obtaining private versions of $S^{(1)}(t,\beta_0)$ and $S^{(0)}(t,\beta_0)$ that…

Let $\calA^t$ denote the set of active clients, let $p_i^t=\mathbb{E}[\Indc_{\{i\in\calA^t\}}]$ be client $i$'s availability probability in round $t$, and consider federated learni…

Let $m$ and $n$ denote the parameters governing the federated conformal prediction setting, let $\beta$ be the target miscoverage level, and let $\text{QQM}$ denote the QQM algorit…

Interactive FDP lower-bound conjecture. The lower bound under the FDP constraint could be strengthened to allow some interaction while the result still holds.

Let $Q$ be a sparsification of a probability distribution $p$, and consider permutation-invariant divergences between $Q$ and $p$. The relevant efficiency result is known for nonne…

In federated learning with inter-agent communication, let the agents' communication network become sufficiently dense. Dense-network conjecture. Server communication rounds might h…

The Quantized Federated TD learning algorithm, or QFedTD, updates the server parameter by … where $\alpha$ is a constant step size, $b_{i,k}$ are independent Bernoulli packet-deliv…

In federated optimization with a small participation rate, let $chat v$ denote the second-moment estimate used by an adaptive method, and consider sign-based compressors that assig…

The recurrent-architecture difficulty conjecture. LSTM-based, or even RNN-based, architectures might be difficult for the proposed dropout score mechanism.

Let CODASCA denote the min-max federated deep AUC maximization algorithm, and suppose it uses the square loss, which is not symmetric: a loss is symmetric when $\ell(z)+\ell(-z)$ i…

Square-root condition-number conjecture. In the more general strongly convex and smooth case, the optimal sample complexity should have a linear dependence on $\sqrt{\kappa}$.

The cited authors consider notions of heterogeneity in federated learning, with the specific notions and the required analyses left unspecified. Heterogeneity benefits conjecture.…

Let $M$ clients play a multi-armed bandit for $T$ time slots. For client $m$, let $k'_{*,m}$ denote its optimal arm, let $X'_{k,m}$ denote the reward distribution of arm $k$ for cl…

Periodic-communication lower-bound conjecture. Unless restrictive assumptions are imposed on the level of statistical or objective heterogeneity, a lower bound of this type should…

Local-step utility conjecture. The phenomenon that increasing $H$ does not improve the convergence rate may be an artifact of a conservative analysis of $\texttt{FedLin}$; a more r…

The setting is simple non-convex optimization in which, for example, a non-convex function $f$ is optimized over a convex set, as tested in the paper's EMNIST experiment. FedDualAv…

COTAF is an over-the-air federated learning scheme based on analog transmissions over a multiple-access channel, and multi-stage federated learning includes clustered federated lea…

Learning-rate trade-off conjecture. Even in much broader settings, the choice of learning rate $gamma$ dictates a trade-off between accuracy and initial convergence.

Learning-rate relaxation conjecture. This condition can be relaxed. The question concerns whether convergence guarantees under a bounded variance assumption can hold with a larger…

Let FedAc denote the federated accelerated stochastic gradient descent method studied in the paper, and let the objective be non-convex. FedAc's non-convex extension conjecture. Fe…

The experiments compare hierarchical federated learning (HFL) and federated learning (FL) on CIFAR-10, including sparse variants, and observe a small degradation in the accuracy of…