Foundation models require fine-tuning to ensure their generative outputs align with intended results for specific tasks. Automating this fine-tuning process is challenging, as it typically needs human feedback that can be expensive to acquire. We present AutoRefine, a method that leverages reinforcement learning for targeted fine-tuning, utilizing direct feedback from measurable performance improvements in specific downstream tasks. We demonstrate the method for a problem arising in algorithmic hiring platforms where linguistic biases influence a recommendation system. In this setting, a generative model seeks to rewrite given job specifications to receive more diverse candidate matches from a recommendation engine which matches jobs to candidates. Our model detects and regulates biases in job descriptions to meet diversity and fairness criteria. The experiments on a public hiring dataset and a real-world hiring platform showcase how large language models can assist in identifying and mitigation biases in the real world.

To this end, we introduce a novel paradigm: Dataset-Agnostic Recommender Systems (DAReS) that aims to enable a single codebase to autonomously adapt to various datasets without the need for fine-tuning, for a given recommender system task. Central to this approach is the Dataset Description Language (DsDL), a structured format that provides metadata about the dataset's features and labels, and allow the system to understand dataset's characteristics, allowing it to autonomously manage processes like feature selection, missing values imputation, noise removal, and hyperparameter optimization. By reducing the need for domainspecific expertise and manual adjustments, DAReS offers a more efficient and scalable solution for building recommender systems across diverse application domains. It addresses critical challenges in the field, such as reusability, reproducibility, and accessibility for non-expert users or entry-level researchers. With DAReS, we hope to spark community's attention in making recommender systems more adaptable, reproducible, and usable, with little to no configuration required from (possibly nonexpert or entry-level) users.

By defining To improve the inference efficiency without sacrificing accuracy, knowledge as different frequency components of the features, we many studies [10, 11, 13, 31] have adopted Knowledge Distillation theoretically demonstrate that regular feature-based knowledge distillation (KD) to recommender system. KD is a model-agnostic is equivalent to equally minimizing losses on all knowledge approach for model compression [6, 8]. In knowledge distillation and further analyze how this equal loss weight allocation method for recommendation, the common process is first to train a large leads to important knowledge being overlooked. In light of this, teacher model using the user-item interactions, then train a small we propose to emphasize important knowledge by redistributing student model using the user-item interactions as well as the features knowledge weights. Furthermore, we propose FreqD, a lightweight in the intermediate layer [10, 11, 13] and the predictions in knowledge reweighting method, to avoid the computational cost the output layer [1, 10, 15, 17] provided by the teacher model.

In business analysis, providing effective recommendations is essential for enhancing company profits. The utilization of graph-based structures, such as bipartite graphs, has gained popularity for their ability to analyze complex data relationships. Link prediction is crucial for recommending specific items to users. Traditional methods in this area often involve identifying patterns in the graph structure or using representational techniques like graph neural networks (GNNs). However, these approaches encounter difficulties as the volume of data increases. To address these challenges, we propose a model called Graph Contrastive Learning for Multi-label Classification (MCGCL). MCGCL leverages contrastive learning to enhance recommendation effectiveness. The model incorporates two training stages: a main task and a subtask. The main task is holistic user-item graph learning to capture user-item relationships. The homogeneous user-user (item-item) subgraph is constructed to capture user-user and item-item relationships in the subtask. We assessed the performance using real-world datasets from Amazon Reviews in multi-label classification tasks. Comparative experiments with state-of-the-art methods confirm the effectiveness of MCGCL, highlighting its potential for improving recommendation systems.

Reinforcement learning (RL) is increasingly being used in the healthcare domain, particularly for the development of personalized health adaptive interventions. Inspired by the success of Large Language Models (LLMs), we are interested in using LLMs to update the RL policy in real time, with the goal of accelerating personalization. We use the text-based user preference to influence the action selection on the fly, in order to immediately incorporate the user preference. We use the term "user preference" as a broad term to refer to a user personal preference, constraint, health status, or a statement expressing like or dislike, etc. Our novel approach is a hybrid method that combines the LLM response and the RL action selection to improve the RL policy. Given an LLM prompt that incorporates the user preference, the LLM acts as a filter in the typical RL action selection. We investigate different prompting strategies and action selection strategies. To evaluate our approach, we implement a simulation environment that generates the text-based user preferences and models the constraints that impact behavioral dynamics. We show that our approach is able to take into account the text-based user preferences, while improving the RL policy, thus improving personalization in adaptive intervention.

Prospective students face the challenging task of selecting a university program that will shape their academic and professional careers. For decision-makers and support services, it is often time-consuming and extremely difficult to match personal interests with suitable programs due to the vast and complex catalogue information available. This paper presents the first information system that provides students with efficient recommendations based on both program content and personal preferences. BERTopic, a powerful topic modeling algorithm, is used that leverages text embedding techniques to generate topic representations. It enables us to mine interest topics from all course descriptions, representing the full body of knowledge taught at the institution. Underpinned by the student's individual choice of topics, a shortlist of the most relevant programs is computed through statistical backtracking in the knowledge map, a novel characterization of the program-course relationship. This approach can be applied to a wide range of educational settings, including professional and vocational training. A case study at a post-secondary school with 80 programs and over 5,000 courses shows that the system provides immediate and effective decision support. The presented interest topics are meaningful, leading to positive effects such as serendipity, personalization, and fairness, as revealed by a qualitative study involving 65 students. Over 98% of users indicated that the recommendations aligned with their interests, and about 94% stated they would use the tool in the future. Quantitative analysis shows the system can be configured to ensure fairness, achieving 98% program coverage while maintaining a personalization score of 0.77. These findings suggest that this real-time, user-centered, data-driven system could improve the program selection process.

In many practical natural language applications, user data are highly sensitive, requiring anonymous uploads of text data from mobile devices to the cloud without user identifiers. However, the absence of user identifiers restricts the ability of cloud-based language models to provide personalized services, which are essential for catering to diverse user needs. The trivial method of replacing an explicit user identifier with a static user embedding as model input still compromises data anonymization. In this work, we propose to let each mobile device maintain a user-specific distribution to dynamically generate user embeddings, thereby breaking the one-to-one mapping between an embedding and a specific user. We further theoretically demonstrate that to prevent the cloud from tracking users via uploaded embeddings, the local distributions of different users should either be derived from a linearly dependent space to avoid identifiability or be close to each other to prevent accurate attribution. Evaluation on both public and industrial datasets using different language models reveals a remarkable improvement in accuracy from incorporating anonymous user embeddings, while preserving real-time inference requirement.

Grok, the AI assistant that's for some reason baked into X, is now available as a standalone app. Like the version that exists as a tab on the social media platform, the Grok app can be used to generate images, summarize text and answer questions, with a conversational tone xAI, the AI assistant's creator, calls "humorous and engaging." The app was first tested with a limited set of users in December 2024, right around the same time X debuted a free tier of Grok that's available to anyone. Prior to that, you needed to pay at least 8 a month for X Premium to have the privilege of using the AI. The limitations of that free access -- 10 requests every two hours, three image analysis request per day -- may also apply to the Grok app.

Some of the world's most popular apps are likely being co-opted by rogue members of the advertising industry to harvest sensitive location data on a massive scale, with that data ending up with a location data company whose subsidiary has previously sold global location data to US law enforcement. The thousands of apps, included in hacked files from location data company Gravy Analytics, include everything from games like Candy Crush and dating apps like Tinder to pregnancy tracking and religious prayer apps across both Android and iOS. Because much of the collection is occurring through the advertising ecosystem--not code developed by the app creators themselves--this data collection is likely happening without users' or even app developers' knowledge. This article was created in partnership with 404 Media, a journalist-owned publication covering how technology impacts humans. "For the first time publicly, we seem to have proof that one of the largest data brokers selling to both commercial and government clients appears to be acquiring their data from the online advertising'bid stream,'" rather than code embedded into the apps themselves, Zach Edwards, senior threat analyst at cybersecurity firm Silent Push and who has followed the location data industry closely, tells 404 Media after reviewing some of the data.

Differentially private (DP) selection involves choosing a high-scoring candidate from a finite candidate pool, where each score depends on a sensitive dataset. This problem arises naturally in a variety of contexts including model selection, hypothesis testing, and within many DP algorithms. Classical methods, such as Report Noisy Max (RNM), assume all candidates' scores are equally sensitive to changes in a single individual's data, but this often isn't the case. To address this, algorithms like the Generalised Exponential Mechanism (GEM) leverage variability in candidate sensitivities. However, we observe that while these algorithms can outperform RNM in some situations, they may underperform in others - they can even perform worse than random selection. In this work, we explore how the distribution of scores and sensitivities impacts DP selection mechanisms. In all settings we study, we find that there exists a mechanism that utilises heterogeneity in the candidate sensitivities that outperforms standard mechanisms like RNM. However, no single mechanism uniformly outperforms RNM. We propose using the correlation between the scores and sensitivities as the basis for deciding which DP selection mechanism to use. Further, we design a slight variant of GEM, modified GEM that generally performs well whenever GEM performs poorly. Relying on the correlation heuristic we propose combined GEM, which adaptively chooses between GEM and modified GEM and outperforms both in polarised settings.