The University of Massachusetts Amherst, founded in 1863, is the flagship campus of the University of Massachusetts system and one of the nation’s top public research universities. Located in an iconic college town, it is the largest public research university in New England, offering over 200 academic programs and is known for its world-class faculty and commitment to diversity. NEC Labs America partners with the University of Massachusetts Amherst to advance speech understanding and natural language reasoning. Our work includes intent prediction and robust conversational AI. Please read about our latest news and collaborative publications with the University of Massachusetts Amherst.

Posts

Redefining Passive in Backscattering with Commodity Devices

/in /by

The recent innovation of frequency-shifted (FS) backscatter allows for backscattering with commodity devices, which are inherently half-duplex. However, their reliance on oscillators for generating the frequency-shifting signal on the tag, forces them to incur the transient phase of the oscillator before steady-state operation. We show how the oscillator’s transient phase can pose a fundamental limitation for battery-less tags, resulting in significantly low bandwidth efficiencies, thereby limiting their practical usage.To this end, we propose a novel approach to FS-backscatter called xSHIFT that shifts the core functionality of FS away from the tag and onto the commodity device, thereby eliminating the need for on-tag oscillators altogether. The key innovation in xSHIFT lies in addressing the formidable challenges that arise in making this vision a reality. Specifically, xSHIFT’s design is built on the construct of beating twin carrier tones through a non-linear device to generate the desired FS signal – while the twin RF carriers are generated externally through a careful embedding into the resource units of commodity WiFi transmissions, the beating is achieved through a carefully-designed passive tag circuitry. We prototype xSHIFT’s tag, which is the same form factor as RFID Gen 2 tags, and characterize its promising real-world performance. We believe xSHIFT demonstrates one of the first, truly passive tag designs that has the potential to bring commodity backscatter to consumer spaces.

Read more

Improving Face Recognition by Clustering Unlabeled Faces in the Wild

/in /by

While deep face recognition has benefited significantly from large-scale labeled data, current research is focused on leveraging unlabeled data to further boost performance, reducing the cost of human annotation. Prior work has mostly been in controlled settings, where the labeled and unlabeled data sets have no overlapping identities by construction. This is not realistic in large-scale face recognition, where one must contend with such overlaps, the frequency of which increases with the volume of data. Ignoring identity overlap leads to significant labeling noise, as data from the same identity is split into multiple clusters. To address this, we propose a novel identity separation method based on extreme value theory. It is formulated as an out-of-distribution detection algorithm, and greatly reduces the problems caused by overlapping-identity label noise. Considering cluster assignments as pseudo-labels, we must also overcome the labeling noise from clustering errors. We propose a modulation of the cosine loss, where the modulation weights correspond to an estimate of clustering uncertainty. Extensive experiments on both controlled and real settings demonstrate our method’s consistent improvements over supervised baselines, e.g., 11.6% improvement on IJB-A verification.

Read more

Active Adversarial Domain Adaptation

/in /by

We propose an active learning approach for transferring representations across domains. Our approach, active adversarial domain adaptation (AADA), explores a duality between two related problems: adversarial domain alignment and importance sampling for adapting models across domains. The former uses a domain discriminative model to align domains, while the latter utilizes the model to weigh samples to account for distribution shifts. Specifically, our importance weight promotes unlabeled samples with large uncertainty in classification and diversity compared to la-beled examples, thus serving as a sample selection scheme for active learning. We show that these two views can be unified in one framework for domain adaptation and transfer learning when the source domain has many labeled examples while the target domain does not. AADA provides significant improvements over fine-tuning based approaches and other sampling methods when the two domains are closely related. Results on challenging domain adaptation tasks such as object detection demonstrate that the advantage over baseline approaches is retained even after hundreds of examples being actively annotated.

Read more