arXiv subjects

Mandana Saebi

Publications and source records attributed to Mandana Saebi.

9 recordsLinked to original sources

arXiv2025

Apple Intelligence Foundation Language Models: Tech Report 2025

We introduce two multilingual, multimodal foundation language models that power Apple Intelligence features across Apple devices and services: i a 3B-parameter on-device model optimized for Apple silicon through architectural innovations such as KV-cache sharing and 2-bit quantization-aware training; and ii a scalable server model built on a novel Parallel-Track Mixture-of-Experts PT-MoE transformer that combines track parallelism, mixture-of-experts sparse computation, and interleaved global-local attention to deliver high quality with competitive cost on Apple's Private Cloud Compute platform. Both models are trained on large-scale multilingual and multimodal datasets sourced via responsible web crawling, licensed corpora, and high-quality synthetic data, then further refined with supervised fine-tuning and reinforcement learning on a new asynchronous platform. The resulting models support several additional languages while understanding images and executing tool calls. In public benchmarks and human evaluations, both the server model and the on-device model match or surpass comparably sized open baselines. A new Swift-centric Foundation Models framework exposes guided generation, constrained tool calling, and LoRA adapter fine-tuning, allowing developers to integrate these capabilities with a few lines of code. The latest advancements in Apple Intelligence models are grounded in our Responsible AI approach with safeguards like content filtering and locale-specific evaluation, as well as our commitment to protecting our users' privacy with innovations like Private Cloud Compute.

cs.LG↗

arXiv2021

Ballast water-mediated species spread risk dynamics and policy implications to reduce the invasion risk to the Mediterranean Sea

The Mediterranean Sea is one of the most heavily invaded marine regions. This work focuses on the dynamics and potential policy options for ballast water-mediated nonindigenous species to the Mediterranean. Specifically, we (1) estimated port risks in years 2012, 2015, and 2018, (2) identified hub ports that connect many clusters, and (3) evaluated four regulatory scenarios. The risk results show that Gibraltar, Suez, and Istanbul remained high-risk ports from 2012-2018, and they served as hub ports that connected several spread clusters. With policy scenario analysis, we found that regulating the high-risk hub ports can disproportionately reduce the overall risk to the Mediterranean: the average risk to all ports was reduced by 5-10% by regulating one high-risk hub port, while the average risk to all ports was only reduced by 0.2% by regulating one average-risk Mediterranean port. We also found that only regulating high-risk ports cannot reduce their risks effectively.

q-bio.PE↗

arXiv2021

Economic and environmental impacts of ballast water management on Small Island Developing States and Least Developed Countries

The Ballast Water Management Convention can decrease the introduction risk of harmful aquatic organisms and pathogens, yet the Convention increases shipping costs and causes subsequent economic impacts. This paper examines whether the Convention generates disproportionate invasion risk reduction results and economic impacts on Small Island Developing States (SIDS) and Least Developed Countries (LDCs). Risk reduction is estimated with an invasion risk assessment model based on a higher-order network, and the effects of the regulation on national economies and trade are estimated with an integrated shipping cost and computable general equilibrium modeling framework. Then we use the Lorenz curve to examine if the regulation generates risk or economic inequality among regions. Risk reduction ratios of all regions (except Singapore) are above 99%, which proves the effectiveness of the Convention. The Gini coefficient of 0.66 shows the inequality in risk changes relative to income levels among regions, but risk reductions across all nations vary without particularly high risks for SIDS and LDCs than for large economies. Similarly, we reveal inequality in economic impacts relative to income levels (the Gini coefficient is 0.58), but there is no evidence that SIDS and LDCs are disproportionately impacted compared to more developed regions. Most changes in GDP, real exports, and real imports of studied regions are minor (smaller than 0.1%). However, there are more noteworthy changes for select sectors and trade partners including Togo, Bangladesh, and Dominican Republic, whose exports may decrease for textiles and metal and chemicals. We conclude the Convention decreases biological invasion risk and does not generate disproportionate negative impacts on SIDS and LDCs.

econ.GN↗

arXiv2021

A Discriminative Entity-Aware Language Model for Virtual Assistants

High-quality automatic speech recognition (ASR) is essential for virtual assistants (VAs) to work well. However, ASR often performs poorly on VA requests containing named entities. In this work, we start from the observation that many ASR errors on named entities are inconsistent with real-world knowledge. We extend previous discriminative n-gram language modeling approaches to incorporate real-world knowledge from a Knowledge Graph (KG), using features that capture entity type-entity and entity-entity relationships. We apply our model through an efficient lattice rescoring process, achieving relative sentence error rate reductions of more than 25% on some synthesized test sets covering less popular entities, with minimal degradation on a uniformly sampled VA test set.

cs.CL↗

arXiv2020

Efficient modeling of higher-order dependencies in networks: from algorithm to application for anomaly detection

Complex systems, represented as dynamic networks, comprise of components that influence each other via direct and/or indirect interactions. Recent research has shown the importance of using Higher-Order Networks (HONs) for modeling and analyzing such complex systems, as the typical Markovian assumption in developing the First Order Network (FON) can be limiting. This higher-order network representation not only creates a more accurate representation of the underlying complex system, but also leads to more accurate network analysis. In this paper, we first present a scalable and accurate model, BuildHON+, for higher-order network representation of data derived from a complex system with various orders of dependencies. Then, we show that this higher-order network representation modeled by BuildHON+ is significantly more accurate in identifying anomalies than FON, demonstrating a need for the higher-order network representation and modeling of complex systems for deriving meaningful conclusions.

cs.SI↗

arXiv2020

Network analysis of ballast-mediated species transfer reveals important introduction and dispersal patterns in the Arctic

Rapid climate change has wide-ranging implications for the Arctic region, including sea ice loss, increased geopolitical attention, and expanding economic activity, including a dramatic increase in shipping activity. As a result, the risk of harmful non-native marine species being introduced into this critical region will increase unless policy and management steps are implemented in response. Using big data about shipping, ecoregions, and environmental conditions, we leverage network analysis and data mining techniques to assess, visualize, and project ballast water-mediated species introductions into the Arctic and dispersal of non-native species within the Arctic. We first identify high-risk connections between the Arctic and non-Arctic ports that could be sources of non-native species over 15 years (1997-2012) and observe the emergence of shipping hubs in the Arctic where the cumulative risk of non-native species introduction is increasing. We then consider how environmental conditions can constrain this Arctic introduction network for species with different physiological limits, thus providing a species-level tool for decision-makers. Next, we focus on within-Arctic ballast-mediated species dispersal where we use higher-order network analysis to identify critical shipping routes that may facilitate species dispersal within the Arctic. The risk assessment and projection framework we propose could inform risk-based assessment and management of ship-borne invasive species in the Arctic.

q-bio.PE↗

arXiv2020

HONEM: Learning Embedding for Higher Order Networks

Representation learning on networks offers a powerful alternative to the oft painstaking process of manual feature engineering, and as a result, has enjoyed considerable success in recent years. However, all the existing representation learning methods are based on the first-order network (FON), that is, the network that only captures the pairwise interactions between the nodes. As a result, these methods may fail to incorporate non-Markovian higher-order dependencies in the network. Thus, the embeddings that are generated may not accurately represent of the underlying phenomena in a network, resulting in inferior performance in different inductive or transductive learning tasks. To address this challenge, this paper presents HONEM, a higher-order network embedding method that captures the non-Markovian higher-order dependencies in a network. HONEM is specifically designed for the higher-order network structure (HON) and outperforms other state-of-the-art methods in node classification, network re-construction, link prediction, and visualization for networks that contain non-Markovian higher-order dependencies.

cs.LG↗

arXiv2020

Heterogeneous Relational Reasoning in Knowledge Graphs with Reinforcement Learning

Path-based relational reasoning over knowledge graphs has become increasingly popular due to a variety of downstream applications such as question answering in dialogue systems, fact prediction, and recommender systems. In recent years, reinforcement learning (RL) has provided solutions that are more interpretable and explainable than other deep learning models. However, these solutions still face several challenges, including large action space for the RL agent and accurate representation of entity neighborhood structure. We address these problems by introducing a type-enhanced RL agent that uses the local neighborhood information for efficient path-based reasoning over knowledge graphs. Our solution uses graph neural network (GNN) for encoding the neighborhood information and utilizes entity types to prune the action space. Experiments on real-world dataset show that our method outperforms state-of-the-art RL methods and discovers more novel paths during the training procedure.

cs.LG↗

arXiv2017

A Minimum-Cost Flow Model for Workload Optimization on Cloud Infrastructure

Recent technology advancements in the areas of compute, storage and networking, along with the increased demand for organizations to cut costs while remaining responsive to increasing service demands have led to the growth in the adoption of cloud computing services. Cloud services provide the promise of improved agility, resiliency, scalability and a lowered Total Cost of Ownership (TCO). This research introduces a framework for minimizing cost and maximizing resource utilization by using an Integer Linear Programming (ILP) approach to optimize the assignment of workloads to servers on Amazon Web Services (AWS) cloud infrastructure. The model is based on the classical minimum-cost flow model, known as the assignment model.

cs.DC↗