We argue that 3-D first-person video games are a challenging environment for real-time multi-modal reasoning. We first describe our dataset of human game-play, collected across a large variety of 3-D first-person games, which is both substantially larger and more diverse compared to prior publicly disclosed datasets, and contains text instructions. We demonstrate that we can learn an inverse dynamics model from this dataset, which allows us to impute actions on a much larger dataset of publicly available videos of human game play that lack recorded actions. We then train a text-conditioned agent for game playing using behavior cloning, with a custom architecture capable of realtime inference on a consumer GPU. We show the resulting model is capable of playing a variety of 3-D games and responding to text input. Finally, we outline some of the remaining challenges such as long-horizon tasks and quantitative evaluation across a large set of games.

Real-world recommender systems often need to balance multiple objectives when deciding which recommendations to present to users. These include behavioural signals (e.g. clicks, shares, dwell time), as well as broader objectives (e.g. diversity, fairness). Scalarisation methods are commonly used to handle this balancing task, where a weighted average of per-objective reward signals determines the final score used for ranking. Naturally, how these weights are computed exactly, is key to success for any online platform. We frame this as a decision-making task, where the scalarisation weights are actions taken to maximise an overall North Star reward (e.g. long-term user retention or growth). We extend existing policy learning methods to the continuous multivariate action domain, proposing to maximise a pessimistic lower bound on the North Star reward that the learnt policy will yield. Typical lower bounds based on normal approximations suffer from insufficient coverage, and we propose an efficient and effective policy-dependent correction for this. We provide guidance to design stochastic data collection policies, as well as highly sensitive reward signals. Empirical observations from simulations, offline and online experiments highlight the efficacy of our deployed approach.

Practitioners who wish to build real-world applications that rely on ranking models, need to decide which modelling paradigm to follow. This is not an easy choice to make, as the research literature on this topic has been shifting in recent years. In particular, whilst Gradient Boosted Decision Trees (GBDTs) have reigned supreme for more than a decade, the flexibility of neural networks has allowed them to catch up, and recent works report accuracy metrics that are on par. Nevertheless, practical systems require considerations beyond mere accuracy metrics to decide on a modelling approach. This work describes our experiences in balancing some of the trade-offs that arise, presenting a case study on a short-video recommendation application. We highlight (1) neural networks' ability to handle large training data size, user- and item-embeddings allows for more accurate models than GBDTs in this setting, and (2) because GBDTs are less reliant on specialised hardware, they can provide an equally accurate model at a lower cost. We believe these findings are of relevance to researchers in both academia and industry, and hope they can inspire practitioners who need to make similar modelling choices in the future.

Recommender systems are an essential component of online platforms providing users with personalized experiences. Some recommendation scenarios such as social networks and news are extremely dynamic in nature with user interests changing over time and new items being continuously added due to breaking news and trending events. Incremental training is a popular technique to keep recommender models up-to-date in such dynamic platforms. In this paper, we provide an empirical analysis of a large industry dataset from the Sharechat app MOJ, a social media platform featuring short videos, to answer relevant questions like - How often should I retrain the models? - do different model architectures, features and dataset sizes benefit differently from incremental training? - Does incremental training equally benefit all users and items?

Listwise ranking losses have been widely studied in recommender systems. However, new paradigms of content consumption present new challenges for ranking methods. In this work we contribute an analysis of learning to rank for personalized mobile push notifications and discuss the unique challenges this presents compared to traditional ranking problems. To address these challenges, we introduce a novel ranking loss based on weighting the pairwise loss between candidates by the expected regret incurred for misordering the pair. We demonstrate that the proposed method can outperform prior methods both in a simulated environment and in a production experiment on a major social network.

Most recommender systems are myopic, that is they optimize based on the immediate response of the user. This may be misaligned with the true objective, such as creating long term user satisfaction. In this work we focus on mobile push notifications, where the long term effects of recommender system decisions can be particularly strong. For example, sending too many or irrelevant notifications may annoy a user and cause them to disable notifications. However, a myopic system will always choose to send a notification since negative effects occur in the future. This is typically mitigated using heuristics. However, heuristics can be hard to reason about or improve, require retuning each time the system is changed, and may be suboptimal. To counter these drawbacks, there is significant interest in recommender systems that optimize directly for long-term value (LTV). Here, we describe a method for maximising LTV by using model-based reinforcement learning (RL) to make decisions about whether to send push notifications. We model the effects of sending a notification on the user’s future behavior. Much of the prior work applying RL to maximise LTV in recommender systems has focused on session-based optimization, while the time horizon for notification decision making in this work extends over several days. We test this approach in an A/B test on a major social network. We show that by optimizing decisions about push notifications we are able to send less notifications and obtain a higher open rate than the baseline system, while generating the same level of user engagement on the platform as the existing, heuristic-based, system.

After the success the RecSys 2020 Challenge, we are describing a novel and bigger dataset that was released in conjunction with the ACM RecSys Challenge 2021. This year's dataset is not only bigger (~ 1B data points, a 5 fold increase), but for the first time it take into consideration fairness aspects of the challenge. Unlike many static datsets, a lot of effort went into making sure that the dataset was synced with the Twitter platform: if a user deleted their content, the same content would be promptly removed from the dataset too. In this paper, we introduce the dataset and challenge, highlighting some of the issues that arise when creating recommender systems at Twitter scale.

A method for developing an enhancement model for visual data, the method comprising the steps of: receiving one or more sections of higher-quality visual data and one or more sections of lower-quality visual data corresponding to the one or more sections of higher-quality data. A hierarchical algorithm is then developed, wherein the hierarchical algorithm is operable to substantially reproduce the one or more sections of higher- quality visual data from the corresponding one or more sections of lower-quality visual data. The developed hierarchical algorithm is then stored in a library of developed hierarchical algorithms.

The workshop features presentations of accepted contributions to the RecSys Challenge 2021, organized by Politecnico di Bari, ETH Zürich, Jönköping University, and the data set is provided by Twitter. The challenge focuses on a real-world task of tweet engagement prediction in a dynamic environment. For 2021, the challenge considers four different engagement types: Likes, Retweet, Quote, and replies. This year’s challenge brings the problem even closer to Twitter’s real recommender systems by introducing latency constraints. We also increases the data size to encourage novel methods. Also, the data density is increased in terms of the graph where users are considered to be nodes and interactions as edges. The goal is twofold: to predict the probability of different engagement types of a target user for a set of Tweets based on heterogeneous input data while providing fair recommendations. In fact, multi-goal optimization considering accuracy and fairness is particularly challenging. However, we believed that the recommendation community was nowadays mature enough to face the challenge of providing accurate and, at the same time, fair recommendations. To this end, Twitter has released a public dataset of close to 1 billion data points, > 40 million each day over 28 days. Week 1 − 3 will be used for training and week 4 for evaluation and testing. Each datapoint contains the tweet along with engagement features, user features, and tweet features. A peculiarity of this challenge is related to keeping the dataset updated with the platform: if a user deletes a Tweet, or their data from Twitter, the dataset is promptly updated. Moreover, each change in the dataset implied new evaluations of all submissions and the update of the leaderboard metrics. The challenge was well received with 578 registered users, and 386 submissions.

A method for enhancing at least a section of lower-quality visual data using a hierarchical algorithm, the method comprises receiving at least one section of lower-quality visual data; and extracting a subset of features, from the at least one section of lower-quality visual data. A plurality of layers of reduced-dimension visual data from the extracted features are formed and enhanced to form at least one section of higher-quality visual data. The at least one section of higher-quality visual data corresponds to the at least one section of lower-quality visual data received.

A system (e.g., an auto-encoder system) includes an encoder, a decoder and a learning module. The encoder generates compressed video data using a lossy compression algorithm, the lossy compression algorithm being implemented using a trained neural network with at least one convolution, generate at least one first parameter based on the compressed video data, and communicate the compressed video data and the model to at least one device configured to decode the compressed video data using an inverse algorithm based on the lossy compression algorithm. The decoder generates decoded video data based on the compressed video data using the inverse algorithm and the model, and generate at least one second parameter based on the decoded video data. The learning module trains the model using the at least one first parameter and the at least one second parameter.

A method for training an algorithm to process at least a section of received visual data using a training dataset and reference dataset. The method comprises an iterative method with iterations comprising: generating a set of training data using the algorithm; comparing one or more characteristics of the training data to one or more characteristics of at least a section of the reference dataset; and modifying one or more parameters of the algorithm to optimise processed visual data based on the comparison between the characteristic of the training data and the characteristic of the reference dataset. The algorithm may output the processed visual data with the same content as the at least a section of received visual data. Some aspects and/or implementations provide for improved super-resolution of lower quality images to produce super-resolution images with improved characteristics (e.g. less blur, less undesired smoothing) compared to other super-resolution techniques.

Systems and methods generate a face-swapped image from a target image using a convolutional neural network trained to apply a source identity to the expression and pose of the target image. The convolutional neural network produces face-swapped images fast enough to transform a video stream. An example method includes aligning the face portion of a target image from an original view to a reference view to generate a target face and generating a swapped face by changing the target face to that of a source identity using a convolutional neural network trained to minimize loss of content from the target face and style from the source identity. The method also includes realigning the swapped face from the reference view to the original view and generating a swapped image by stitching the realigned swapped face with the remaining portion of the target image.

Filter bank multicarrier with offset quadrature amplitude modulation (FBMC/OQAM) is considered as a powerful supplementary waveform for future wireless communications. However, FBMC systems have the same high peak-to-average power ratio (PAPR) problem as other multi-carrier systems, and the PAPR reduction methods designed for orthogonal frequency division multiplexing (OFDM) systems cannot be directly applied to FBMC systems due to the unique overlapping structure of FBMC signals. Therefore, some PAPR suppression schemes tailored to FBMC systems have been proposed, but their computational complexity is prohibitively high for some practical applications. In this paper, we propose a conversion vector-based low-complexity dispersive selection mapping (DSLM) scheme, called the C-DSLM scheme, to reduce the PAPR of FBMC signals. In the C-DSLM scheme, a series of conversion vectors is first carefully designed for FBMC signals with overlapping structures, and then candidate signals are generated by multiplying the original signal by the cyclic shift of the conversion vectors with only a few nonzero elements. The complexity evaluation and simulation results show that the C-DSLM scheme's PAPR suppression performance is similar to that of the DSLM scheme with computational complexity is only approximately 10% of that of the DSLM scheme. Our proposed C-DSLM scheme is highly efficient and strongly applicable.

The workshop features presentations of accepted contributions to the RecSys Challenge 2020, organized by Politecnico di Bari, Free University of Bozen-Bolzano, TU Wien, University of Colorado, Boulder, and Universidade Federal de Campina Grande, and sponsored by Twitter. The challenge focuses on a real-world task of Tweet engagement prediction in a dynamic environment. The goal is to predict the probability for different types of engagement (Like, Reply, Retweet, and Retweet with comment) of a target user for a set of Tweets, based on heterogeneous input data. To this end, Twitter has released a large public dataset of ~160M public Tweets, obtained by subsampling within ~2 weeks, that contains engagement features, user features, and Tweet features. A peculiarity of this challenge is related to the recent regulations on data protection and privacy. The challenge data set was compliant: if a user deleted a Tweet, or their data from Twitter, the dataset was promptly updated. Moreover, each change in the dataset implied new evaluations of all submissions and the update of the leaderboard metrics. The challenge was well received with 1,131 registered users. In the final phase, 20 teams were competing for the winning position. These teams had an average size of approximately 4 participants and developed an overall number of 127 different methods.

Recommender systems constitute the core engine of most social network platforms nowadays, aiming to maximize user satisfaction along with other key business objectives. Twitter is no exception. Despite the fact that Twitter data has been extensively used to understand socioeconomic and political phenomena and user behaviour, the implicit feedback provided by users on Tweets through their engagements on the Home Timeline has only been explored to a limited extent. At the same time, there is a lack of large-scale public social network datasets that would enable the scientific community to both benchmark and build more powerful and comprehensive models that tailor content to user interests. By releasing an original dataset of 160 million Tweets along with engagement information, Twitter aims to address exactly that. During this release, special attention is drawn on maintaining compliance with existing privacy laws. Apart from user privacy, this paper touches on the key challenges faced by researchers and professionals striving to predict user engagements. It further describes the key aspects of the RecSys 2020 Challenge that was organized by ACM RecSys in partnership with Twitter using this dataset.

Deep Neural Networks (DNNs) with sparse input features have been widely used in recommender systems in industry. These models have large memory requirements and need a huge amount of training data. The large model size usually entails a cost, in the range of millions of dollars, for storage and communication with the inference services. In this paper, we propose a hybrid hashing method to combine frequency hashing and double hashing techniques for model size reduction, without compromising performance. We evaluate the proposed models on two product surfaces. In both cases, experiment results demonstrated that we can reduce the model size by around 90 % while keeping the performance on par with the original baselines.

Recommender systems trained in a continuous learning fashion are plagued by the feedback loop problem, also known as algorithmic bias. This causes a newly trained model to act greedily and favor items that have already been engaged by users. This behavior is particularly harmful in personalised ads recommendations, as it can also cause new campaigns to remain unexplored. Exploration aims to address this limitation by providing new information about the environment, which encompasses user preference, and can lead to higher long-term reward. In this work, we formulate a display advertising recommender as a contextual bandit and implement exploration techniques that require sampling from the posterior distribution of click-through-rates in a computationally tractable manner. Traditional large-scale deep learning models do not provide uncertainty estimates by default. We approximate these uncertainty measurements of the predictions by employing a bootstrapped model with multiple heads and dropout units. We benchmark a number of different models in an offline simulation environment using a publicly available dataset of user-ads engagements. We test our proposed deep Bayesian bandits algorithm in the offline simulation and online AB setting with large-scale production traffic, where we demonstrate a positive gain of our exploration model.