In many real-world hybrid (mixed discrete continuous) planning problems such as Reservoir Control, Heating, Ventilation and Air Conditioning (HVAC), Navigation, it is difficult to obtain a model of the complex nonlinear dynamics that govern state evolution. However, ubiquity modern sensors allow us collect large quantities data from each these systems build accurate, deep network models their transitions. But there remains one major problem for task control -- how can we plan with learned...

Critiquing is a method for conversational recommendation that adapts recommendations in response to user preference feedback regarding item attributes. Historical critiquing methods were largely based on constraint- and utility-based modifying w.r.t. these critiqued In this paper, we revisit the approach from lens of deep learning language-based interaction. Concretely, propose an end-to-end framework with two variants extend Neural Collaborative Filtering architecture explanation...

Previous highly scalable One-Class Collaborative Filtering (OC-CF) methods such as Projected Linear Recommendation (PLRec) have advocated using fast randomized SVD to embed items into a latent space, followed by linear regression learn personalized recommendation models per user. However, naive embedding often exhibit strong popularity bias that prevents them from accurately less popular items, which is exacerbated the extreme sparsity of implicit feedback matrices in OC-CF setting. To...

Critiquing is a method for conversational recommendation that iteratively adapts recommendations in response to user preference feedback. In this setting, provided with an item and attribute description item; may either accept the recommendation, or critique attributes generate new recommendation. Historical critiquing methods were largely based on explicit constraint- utility-based modifying w.r.t. critiqued attributes. paper, we revisit approach era of latent embeddings subjective...

Preserving the performance of a trained model while removing unique characteristics marked training data points is challenging. Recent research usually suggests retraining from scratch with remaining or refining by reverting optimization on points. Unfortunately, aside their computational inefficiency, those approaches inevitably hurt resulting model's generalization ability since they remove not only but also discard shared (and possibly contributive) information. To address degradation...

Providing explanations for recommended items not only allows users to understand the reason receiving recommendations but also provides with an opportunity refine by critiquing undesired parts of explanation. While much research focuses on improving explanation recommendations, less effort has focused interactive recommendation allowing a user critique explanations. Aside from traditional constraint- and utility-based systems, end-to-end deep learning based approach in literature so far,...

Automatic playlist continuation is a prominent problem in music recommendation. Significant portion of consumption now done online through playlists and playlist-like radio stations. Manually compiling for consumers highly time consuming task that difficult to do at scale given the diversity tastes large amount musical content available. Consequently, automated has received increasing attention recently [1, 7, 11]. The 2018 ACM RecSys Challenge [14] dedicated evaluating advancing current...

Critiquing is a method for conversational recommendation that incrementally adapts recommendations in response to user preference feedback. Specifically, iteratively provided with item and attribute descriptions those items; the may then either accept or choose critique an generate new recommendation. A recent direction known as latent linear critiquing (LLC) takes modern embedding-based approach seeks optimize combination of embeddings critiques based on subjective (i.e., keyphrases from...

Social media platforms such as Twitter or StockTwits are widely used for sharing stock market opinions between investors, traders, and entrepreneurs. Empirically, previous work has shown that the content posted on these social can be leveraged to predict various aspects of performance. Nonetheless, actors may not always have altruistic motivations instead seek influence trading behavior through (potentially misleading) information they post. While a lot sought analyze how market, there...

Given recent deep learning results that demonstrate the ability to effectively optimize high-dimensional non-convex functions with gradient descent optimization on GPUs, we ask in this paper whether symbolic tools such as Tensorflow can be effective for planning hybrid (mixed discrete and continuous) nonlinear domains high dimensional state action spaces? To end, RMSProp is competitive mixed integer linear program (MILP) based piecewise (where compute optimal solutions) substantially...

In many complex planning problems with factored, continuous state and action spaces such as Reservoir Control, Heating Ventilation Air Conditioning (HVAC), Navigation domains, it is difficult to obtain a model of the nonlinear dynamics that govern evolution. However, ubiquity modern sensors allows us collect large quantities data from each these systems build accurate, deep neural network models their transitions. But there remains one major problem for task control – how can we plan learned...

Variational Autoencoder (VAE) based methods for Collaborative Filtering (CF) demonstrate remarkable performance one-class (implicit negative) recommendation tasks by extending autoencoders with relaxed but tractable latent distributions. Explicitly modeling a distribution over user preferences allows VAEs to learn and item representations that not only reproduce observed interactions, also generalize them leveraging learning from similar users items. Unfortunately, VAE-CF can exhibit...

Self-supervised representation learning~(SSRL) has advanced considerably by exploiting the transformation invariance assumption under artificially designed data augmentations. While augmentation-based SSRL algorithms push boundaries of performance in computer vision and natural language processing, they are often not directly applicable to other modalities, can conflict with application-specific augmentation constraints. This paper presents an approach that be applied any modality network...

Preserving the performance of a trained model while removing unique characteristics marked training data points is challenging. Recent research usually suggests retraining from scratch with remaining or refining by reverting optimization on points. Unfortunately, aside their computational inefficiency, those approaches inevitably hurt resulting model's generalization ability since they remove not only but also discard shared (and possibly contributive) information. To address degradation...

Interactive (a.k.a. conversational) recommendation systems provide the potential capability to personalize interactions with increasingly prevalent dialog-based AI assistants. In conversational setting, a user often has long-term preferences inferred from previous along ephemeral session-based that need be efficiently elicited through minimal interaction. Historically, Bayesian preference elicitation methods have proved effective for (i) leveraging prior information incrementally estimate...

In many real-world planning problems with factored, mixed discrete and continuous state action spaces such as Reservoir Control, Heating Ventilation, Air Conditioning, Navigation domains, it is difficult to obtain a model of the complex nonlinear dynamics that govern evolution. However, ubiquity modern sensors allows us collect large quantities data from each these systems build accurate, deep neural network models their transitions. But there remains one major problem for task control --...

Within-basket recommendation (WBR) refers to the task of recommending items end completing a non-empty shopping basket during session. While latest innovations in this space demonstrate remarkable performance improvement on benchmark datasets, they often overlook complexity user behaviors practice, such as 1) co-existence multiple intentions, 2) multi-granularity and 3) interleaving behavior (switching intentions) This paper presents Neural Pattern Associator (NPA), deep...

Existing example-based prediction explanation methods often bridge test and training data points through the model's parameters or latent representations. While these offer clues to causes of model predictions, they exhibit innate shortcomings, such as incurring significant computational overhead producing coarse-grained explanations. This paper presents a Highly-precise Data-centric Explanation (HD-Explain), straightforward method exploiting properties Kernelized Stein Discrepancy (KSD)....

Weight-preserving model editing techniques heavily rely on the scoping mechanism that decides when to apply an edit base model. These mechanisms utilize distance functions in representation space ascertain scope of edit. In this work, we show distance-based grapple with lexical biases leading issues such as misfires irrelevant prompts share similar characteristics. To address problem, introduce, Projector Editor Networks for Model Editing (PENME),is a approach employs compact adapter...

In light of recent privacy regulations, machine unlearning has attracted significant attention in the research community. However, current studies predominantly assess overall success approaches, overlooking varying difficulty individual training samples. As a result, broader feasibility remains under-explored. This paper presents set novel metrics for quantifying by jointly considering properties target model and data distribution. Specifically, we propose several heuristics to conditions...