Search

Tasks critical to enterprise profitability, such as customer churn prediction, fraudulent account detection or customer lifetime value estimation, are often tackled by models trained on features engineered from customer data in tabular format. Application-specific feature engineering adds development, operationalization and maintenance costs over time. Recent advances in representation learning present an opportunity to simplify and generalize feature engineering across applications. When applying these advancements to tabular data researchers deal with data heterogeneity, variations in customer engagement history or the sheer volume of enterprise datasets. In this paper, we propose a novel approach to encode tabular data containing customer transactions, purchase history and other interactions into a generic representation of a customer's association with the business. We then evaluate these embeddings as features to train multiple models spanning a variety of applications. CASPR, Customer Activity Sequence-based Prediction and Representation, applies Transformer architecture to encode activity sequences to improve model performance and avoid bespoke feature engineering across applications. Our experiments at scale validate CASPR for both small and large enterprise applications., Comment: Presented at the Table Representation Learning Workshop, NeurIPS 2022, New Orleans. Authors listed in random order

Federated graph representation learning (FedGRL) brings the benefits of distributed training to graph structured data while simultaneously addressing some privacy and compliance concerns related to data curation. However, several interesting real-world graph data characteristics viz. label deficiency and downstream task heterogeneity are not taken into consideration in current FedGRL setups. In this paper, we consider a realistic and novel problem setting, wherein cross-silo clients have access to vast amounts of unlabeled data with limited or no labeled data and additionally have diverse downstream class label domains. We then propose a novel FedGRL formulation based on model interpolation where we aim to learn a shared global model that is optimized collaboratively using a self-supervised objective and gets downstream task supervision through local client models. We provide a specific instantiation of our general formulation using BGRL a SoTA self-supervised graph representation learning method and we empirically verify its effectiveness through realistic cross-slio datasets: (1) we adapt the Twitch Gamer Network which naturally simulates a cross-geo scenario and show that our formulation can provide consistent and avg. 6.1% gains over traditional supervised federated learning objectives and on avg. 1.7% gains compared to individual client specific self-supervised training and (2) we construct and introduce a new cross-silo dataset called Amazon Co-purchase Networks that have both the characteristics of the motivated problem setting. And, we witness on avg. 11.5% gains over traditional supervised federated learning and on avg. 1.9% gains over individually trained self-supervised models. Both experimental results point to the effectiveness of our proposed formulation. Finally, both our novel problem setting and dataset contributions provide new avenues for the research in FedGRL., Comment: FedGraph'22 workshop (non archival) version. (https://sites.google.com/view/fedgraph2022/accepted-papers)

In this paper, we consider a bandit problem in which there are a number of groups each consisting of infinitely many arms. Whenever a new arm is requested from a given group, its mean reward is drawn from an unknown reservoir distribution (different for each group), and the uncertainty in the arm's mean reward can only be reduced via subsequent pulls of the arm. The goal is to identify the infinite-arm group whose reservoir distribution has the highest $(1-\alpha)$-quantile (e.g., median if $\alpha = \frac{1}{2}$), using as few total arm pulls as possible. We introduce a two-step algorithm that first requests a fixed number of arms from each group and then runs a finite-arm grouped max-quantile bandit algorithm. We characterize both the instance-dependent and worst-case regret, and provide a matching lower bound for the latter, while discussing various strengths, weaknesses, algorithmic improvements, and potential lower bounds associated with our instance-dependent upper bounds., Comment: ALT 2023

In this work, we unveil the effect of RS, induced by a current-voltage hysteresis cycles across CVD-grown monolayer MoS2 based gated RRAM, on its transistors electrical and reliability characteristics. A unique gate voltage dependence on the RS is identified which has a remarkable impact on the switching performance of MoS2 RRAM. RS behavior was found to be significantly dependent on the charge conduction in the channel. Moreover, we have shown a potential device forming event when MoS2 gated RRAMs were subjected to a steady-state electrical stress. Both hysteresis and steady state electrical stress were found to disturb the transistor action of these gated RRAMs, which in fact can be used as a signature of RS. Interestingly, current-voltage hysteresis resulted in unipolar RS, whereas steady-state electrical stress before RS measurement led to bipolar RS. Moreover, successive stress cycles of such an electrical stress leads to multiple resistance states, a behavior similar to synaptic properties like long-term potentiation and long-term depression, typically found in memristors. We find that charge transport mechanism dominant in the MoS2 FET in conjunction with steady-state stress induced device forming determine the extent of RS induced in these MoS2 based gated RRAMs. Finally on the basis of insights developed from the dependence on charge transport mechanism and steady-state stress induced forming of MoS2 channel, we propose a certain steady-state electrical stress condition which can be used as a forming process employed prior to use of MoS2 based binary RRAMs for improved switching performance.

We report, for CVD-grown monolayer MoS2, the very first results on temporal degradation of material and device performance under electrical stress. Both low and high field regimes of operation are explored at different temperatures, gate bias and stress cycles. During low field operation, current is found to saturate after hundreds of seconds of operation with the current decay time constant being a function of temperature and stress cycle. Current saturation after several seconds during low field operation occurs when a thermal equilibrium is established. However, high field operation, especially at low temperature, leads to impact ionization assisted material and device degradation. It is found that high field operation at low temperature results in amorphization of the channel and is verified by device and Kelvin Probe Force Microscopy (KPFM) analyses. In general, a prolonged room temperature operation of CVD-grown MoS2 transistors lead to degraded gate control, higher OFF state current and negative shift in threshold voltage (VT). This is further verified, through micro-Raman and Photoluminescence spectroscopy, which suggest that a steady state DC electrical stress leads to the formation of localized low resistance regions in the channel and a subsequent loss of transistor characteristics. Our findings unveil unique mechanism by which CVD MoS2 undergoes material degradation under electrical stress and subsequent breakdown of transistor behavior. Such an understanding of material and device reliability helps in determining the safe operating regime from device as well as circuit perspective.

Unlike Si, 2-dimensional (2D) Transition Metal Dichalcogenides (TMDs) offer atomically thin channels for carrier transport in FETs. Despite advantages like superior gate control, steep sub-threshold swing and high carrier mobility offered by 2D FET channels, process related challenges like lack of selective doping techniques like implantation and CMOS compatible process for fabrication of 2D TMD based FETs hinder the anticipated viability of 2D semiconductor technology for future electronic applications. In this letter, we demonstrate a process oriented approach to realize superior ambipolarity in 2D FETs based on TMDs like Molybdenum disulfide (MoS2), Tungsten disulfide (WS2), Molybdenum diselenide (MoSe2) and Tungsten diselenide (WSe2) by enhancing hole current by multiple orders of magnitude in otherwise strong N-type transistors. The method involves Hydrogen Sulfide (H2S gas) assisted contact engineering of N-type FETs to introduce surface states that alter device behavior. Based on material characterization and bandstructure calculations, physical insights have been developed to understand the effect of such a contact engineering technique. Subsequently, this technique has been demonstrated to alter device behavior by enhancing hole conduction in originally N-type exfoliated (MoS2, WS2, MoSe2 and WSe2) and CVD grown (MoS2) TMD samples to confirm its potential towards enabling the feasibility of 2D semiconductor device technology.

Metal-semiconductor interface is a bottleneck for efficient transport of charge carriers through Transition Metal Dichalcogenide (TMD) based field-effect transistors (FETs). Injection of charge carriers across such interfaces is mostly limited by Schottky barrier at the contacts which must be reduced to achieve highly efficient contacts for carrier injection into the channel. Here we introduce a universal approach involving dry chemistry to enhance atomic orbital interaction between various TMDs (MoS2, WS2, MoSe2 and WSe2) & metal contacts has been experimentally demonstrated. Quantum chemistry between TMDs, Chalcogens and metals has been explored using detailed atomistic (DFT & NEGF) simulations, which is then verified using Raman, PL and XPS investigations. Atomistic investigations revealed lower contact resistance due to enhanced orbital interaction and unique physics of charge sharing between constituent atoms in TMDs with introduced Chalcogen atoms which is subsequently validated through experiments. Besides contact engineering, which lowered contact resistance by 72, 86, 1.8, 13 times in MoS2, WS2, MoSe2 and WSe2 respectively, a novel approach to cure / passivate dangling bonds present at the 2D TMD channel surface has been demonstrated. While the contact engineering improved the ON-state performance (ION, gm, mobility and RON) of 2D TMD FETs by orders of magnitude, Chalcogen based channel passivation was found to improve gate control (IOFF, SS, & VTH) significantly. This resulted in an overall performance boost. The engineered TMD FETs were shown to have performance on par with best reported till date.

One of the major roadblocks for the establishment of 2D semiconductor technology for CMOS integrated circuits is lack of industry scalable doping techniques that lead to 2D FETs with comparable n-type and p-type behavior. Here we demonstrate a Chalcogen based technique to alter the surface of WSe2 to realize enhanced ambipolar behavior along with a complete transition from n to p-type behavior of WSe2 FETs. The technique involves dry chemistry between Chalcogen atom and TMDC surface which leads to surface states that cause improved hole and electron injection through the FETs. We propose such a technique for realization of all WSe2 based CMOS integrated circuits and therefore unveil its potential towards technology.