Smaller yet useful tech accessories -- batteries, cables, phone cases and so on -- are what keep the big ticket items going. And Amazon's Prime Day is a good time to stock up on them. Only problem is, not everything on Amazon's site passes muster, and the cheaper stuff can be particularly questionable. Luckily, we've tested plenty of these devices for one buyer's guide or another, so we think stuff picked from this list should serve you well. Here are the best Prime Day Tech deals under 50.

When it comes to speakers, you often get what you pay for -- which makes Amazon's second Prime Day of the year a very good time to pick up highly rated (and otherwise pricey) Bluetooth or smart speakers while they're cheaper than usual. Of course, there are thousands of speakers in Amazon's inventory, and not all of them are winners. Here, we've rounded up all the Prime Day speaker deals on the best speakers we've tested, reviewed and currently recommend. Whether you just want some tunes as you horbgorble around at home or need to entertain some coworkers at a music dance experience, there's something for you here -- and best of all, these are on sale. Tribit StormBox Micro 2 for 48 ( 32 off): This is the smallest music box on our list, and we like it because it packs serious sound for its size. The audio isn't the highest fidelity, but the rubbery strap is perfect for strapping onto a pack, bike handlebar or elsewhere.

If you're looking for an affordable Echo speaker to add Alexa to another room in your home, this Prime Day deal on the Echo Dot will be hard to beat. The Echo Dot (5th gen) has dropped to 23 for Prime Day, which is cheaper than it was during the July sales event. This tiny smart speaker has improved audio that competes with more expensive rivals like the HomePod mini. This Echo Dot model launched in 2022 with clearer vocals, deeper bass and more vibrant overall sound than previous generations. Save big on the 2022 Echo Dot.

Summary: This paper introduces an exact algorithm for greedy mode finding for DPPs which is faster by a factor of M (ground set size) than previous work on greedy MAP algorithms for DPPs; the authors also show that this algorithm can be further sped up when diversity is required over only a sliding window within long recommendations. As an additional contribution, the authors show that modeling recommendation problems with DPPs and generating recommendations via their algorithm outperforms other standard (non-DPP) recommender algorithms along various metrics. As the authors mention, a key advantage of DPPs is their ability to tractably balance quality and diversity requirements for most operations, with mode estimation being one of the only operations that remains NP-hard. Indeed, sampling from a DPP has been used in previous literature, presumably as a more scalable alternative to greedy MAP finding (e.g. for network compression). Although the usefulness of DPPs for recommender systems is now an accepted fact, the analysis provided in section 5 and 6.2 remains interesting, in particular thanks to the discussion of the tunable scaling of diversity and quality preferences and how it can easily be incorporated into the new formulation of the greedy algorithm.

Summary: This work considers learning user preferences using a bandit model. The reward is not only based on the judgement of the user, but also whether the user examined the arm. That is feedback examination * judgement In particular, if a user does not examine an arm, lack of feedback does not necessarily indicate that the user does not "like" the arm. This work uses a latent model for the (unobserved) examination of arms, and posits that the probability of positive feedback (binary) can be expressed as a product of the probability of examination (logistic) and positive feedback (logistic). The work proposes a VI approach to estimating the parameters, and then use a Thompson Sampling approach from the approximate posterior as policy. This allows them to use machinery from Russo and Van Roy to obtain regret bounds.

Paper revolves over the observation that in e-commerce world customers rarely purchase two items that belong to the same category (e.g. Therefore, they claim that a robust recommendation system should incorporate both utility and time utility. An additional problem that is tackled in the paper is that many e-commerce systems have no explicit negative feedback to learn from (for example one can see only what items customer purchased - positives - and no explicit negatives in form of items user did not like). I believe that the second problem they mention is not as big of a concern as advertised by authors. In absence of any explicit negative signal good replacements are long dwell-time clicks that did not end up in purchase, as well as cart additions that did not end up in the final purchase or returns. Many companies are also implementing swipe to dismiss that is useful for collecting explicit negative signal and can be applied to any e-commerce site easily.

Perhaps it should be mentioned that such results originate from the normal SBM where both the information-theoretic threshold for detection, and the conjectured algorithmic threshold were studied in detail, e.g. in "Asymptotic analysis of the stochastic block model for modular networks and its algorithmic applications" by Decelle et al. Also in that case the gap between the two threshold is d (for large d). While the main contribution of the paper is theoretical, it would have been nice to see some practical demonstration of the algorithm, comparison to other algorithms (at the same time this should not be used as an argument for rejection). Evidence of the scalability of the algorithm should be presented. Minor points: While the o(), O(), \Omega() notations are rather standard I was not very familiar with the \omega() and had to look it up to be sure. Perhaps more of NIPS audience would not be familiar with those and the definition could be shortly reminded. I've read the author's feedback and took it into account in my score.

In Reinforcement Learning (RL), multi-armed Bandit (MAB) problems have found applications across diverse domains such as recommender systems, healthcare, and finance. Traditional MAB algorithms typically assume stationary reward distributions, which limits their effectiveness in real-world scenarios characterized by non-stationary dynamics. This paper addresses this limitation by introducing and evaluating novel Bandit algorithms designed for non-stationary environments. First, we present the Adaptive Discounted Thompson Sampling (ADTS) algorithm, which enhances adaptability through relaxed discounting and sliding window mechanisms to better respond to changes in reward distributions. We then extend this approach to the Portfolio Optimization problem by introducing the Combinatorial Adaptive Discounted Thompson Sampling (CADTS) algorithm, which addresses computational challenges within Combinatorial Bandits and improves dynamic asset allocation. Additionally, we propose a novel architecture called Bandit Networks, which integrates the outputs of ADTS and CADTS, thereby mitigating computational limitations in stock selection. Through extensive experiments using real financial market data, we demonstrate the potential of these algorithms and architectures in adapting to dynamic environments and optimizing decision-making processes. For instance, the proposed bandit network instances present superior performance when compared to classic portfolio optimization approaches, such as capital asset pricing model, equal weights, risk parity, and Markovitz, with the best network presenting an out-of-sample Sharpe Ratio 20% higher than the best performing classical model.

This study introduces CUPID, a novel approach to session-based reciprocal recommendation systems designed for a real-time one-on-one social discovery platform. In such platforms, low latency is critical to enhance user experiences. However, conventional session-based approaches struggle with high latency due to the demands of modeling sequential user behavior for each recommendation process. Additionally, given the reciprocal nature of the platform, where users act as items for each other, training recommendation models on large-scale datasets is computationally prohibitive using conventional methods. To address these challenges, CUPID decouples the time-intensive user session modeling from the real-time user matching process to reduce inference time. Furthermore, CUPID employs a two-phase training strategy that separates the training of embedding and prediction layers, significantly reducing the computational burden by decreasing the number of sequential model inferences by several hundredfold. Extensive experiments on large-scale Azar datasets demonstrate CUPID's effectiveness in a real-world production environment. Notably, CUPID reduces response latency by more than 76% compared to non-asynchronous systems, while significantly improving user engagement.

Mutual funds aim to generate returns above market averages. While predicting their future portfolio allocations can bring economic advantages, the task remains challenging and largely unexplored. To fill that gap, this work frames mutual fund portfolio prediction as a Next Novel Basket Recommendation (NNBR) task, focusing on predicting novel items in a fund's next portfolio. We create a comprehensive benchmark dataset using publicly available data and evaluate the performance of various recommender system models on the NNBR task. Our findings reveal that predicting novel items in mutual fund portfolios is inherently more challenging than predicting the entire portfolio or only repeated items. While state-of-the-art NBR models are outperformed by simple heuristics when considering both novel and repeated items together, autoencoder-based approaches demonstrate superior performance in predicting only new items. The insights gained from this study highlight the importance of considering domain-specific characteristics when applying recommender systems to mutual fund portfolio prediction. The performance gap between predicting the entire portfolio or repeated items and predicting novel items underscores the complexity of the NNBR task in this domain and the need for continued research to develop more robust and adaptable models for this critical financial application.