Maha Issa

Jihan Ghanim, Maha Issa, Mariette Awad

Building an accurate load forecasting model with minimal underpredictions is vital to prevent any undesired power outages due to underproduction of electricity. However, the power consumption patterns of the residential sector contain fluctuations and anomalies making them challenging to predict. In this paper, we propose multiple Long Short-Term Memory (LSTM) frameworks with different asymmetric loss functions to impose a higher penalty on underpredictions. We also apply a density-based spatial clustering of applications with noise (DBSCAN) anomaly detection approach, prior to the load forecasting task, to remove any present oultiers. Considering the effect of weather and social factors, seasonality splitting is performed on the three considered datasets from France, Germany, and Hungary containing hourly power consumption, weather, and calendar features. Root-mean-square error (RMSE) results show that removing the anomalies efficiently reduces the underestimation and overestimation errors in all the seasonal datasets. Additionally, asymmetric loss functions and seasonality splitting effectively minimize underestimations despite increasing the overestimation error to some degree. Reducing underpredictions of electricity consumption is essential to prevent power outages that can be damaging to the community.

Maha Issa, Anoosheh Heidarzadeh

We introduce the \emph{Private Structured-Subset Retrieval (PSSR)} problem, where a user retrieves $D$ messages from a database of $K$ messages replicated across $N$ non-colluding servers, and the demand is restricted to a known structured family of $D$-subsets. This formulation generalizes classical Private Information Retrieval (PIR) and multi-message PIR (MPIR), and captures settings where the demand space is constrained by application-specific structure. Focusing on balanced ${\{0,1\}}$-linear schemes, we derive converse bounds on the maximum retrieval rate and minimum subpacketization level, and develop an optimization-based framework for constructing schemes for general structured demand families. Our results show that, for certain families, the PSSR rate converse bound can exceed the best-known MPIR rate upper bound; when this PSSR bound is achievable, MPIR rate-optimal schemes become suboptimal for those families. By exploiting demand structure, our PSSR schemes achieve higher retrieval rates for many families and never underperform the best-known balanced ${\{0,1\}}$-linear MPIR schemes. Our results also show that demand structure can reduce the required subpacketization even when the optimal rate is unchanged. Our parallel work on contiguous-demand families further illustrates the scope of this framework by yielding rate-optimal schemes with substantially smaller subpacketization and no field-size restrictions, improving upon MPIR-based schemes.

Maha Issa, Anoosheh Heidarzadeh

This work introduces the \emph{Secure and Private Structured-Subset Retrieval (SPSSR)} problem. In SPSSR, a user wishes to retrieve one subset from an arbitrary family of size-$D$ subsets from $K$ messages replicated across $N$ non-colluding servers that share randomness unknown to the user. The privacy requirement ensures that no server learns which subset is requested, while the security requirement ensures that the user learns nothing about the messages outside the requested subset. This generalizes Symmetric Multi-message Private Information Retrieval (SMPIR), where the candidate demand sets consist of all size-$D$ subsets. We show that, for every candidate demand family, the maximum achievable retrieval rate is equal to ${1-1/N}$. We also show that the minimum ratio between the size of the shared randomness and the message size required to achieve this rate is ${D/(N-1)}$, and that, for balanced linear SPSSR schemes, the minimum required subpacketization level is ${(N-1)/\gcd(D,N-1)}$; both quantities are independent of the demand family. Our converse proof for the maximum achievable retrieval rate applies to arbitrary demand families, unlike the existing proof for SMPIR, which is tailored to the full demand family. For achievability, we construct a single SPSSR scheme that applies uniformly to every demand family, achieves the optimal retrieval rate with the optimal shared-randomness ratio, and requires the optimal subpacketization level among balanced linear schemes. This subpacketization level is no larger than that of known SMPIR schemes in any parameter regime and is smaller in some regimes.

Maha Issa, Anoosheh Heidarzadeh

We introduce the \emph{Private Contiguous-Block Retrieval (PCBR)} problem, where a user retrieves a block of $D$ messages with contiguous indices from $K$ replicated messages stored across $N$ non-colluding servers, while hiding the identity of the requested block from each server. This problem is motivated by storage and streaming systems where files are split into ordered segments. Unlike multi-message Private Information Retrieval (MPIR), where any $D$-subset may be requested, PCBR restricts the demand family to contiguous blocks. This relaxation raises a natural question: Can this structure be exploited to improve retrieval efficiency? We answer this question for balanced $\{0,1\}$-linear schemes. We establish an upper bound on the achievable retrieval rate for all problem parameters, derive a lower bound on the subpacketization level required by any scheme achieving the rate upper bound, and construct a rate-optimal scheme whose subpacketization level matches the lower bound for a broad range of problem parameters. Although the optimal PCBR rate coincides with the best-known MPIR rate converse bound, existing MPIR schemes can be suboptimal for PCBR and can require a much larger subpacketization level. In contrast, our scheme exploits the contiguous-block structure to achieve the optimal rate with reduced subpacketization.