S. Narayanan

Joao Araujo, Alex Chow, Mark Handley et al.

Tail latency dominates the performance of synchronous pretraining jobs when running at very large scales. We describe a three-pronged approach: (1) a new RDMA-based transport protocol, MRC, sprays across many paths and actively load-balances between them, eliminating the issue of flow collisions (2) the use of multi-plane Clos topologies to get the benefits of high switch radix and redundancy, allowing training clusters well over 100K GPUs to be built as two-tier topologies while increasing physical redundancy, and (3) the use of static source-routing using SRv6 to allow MRC the freedom to bypass failures by itself. We describe our experiences running MRC and static SRv6 routing in production in OpenAI and Microsoft's largest training clusters, where it has been used to train the latest frontier models. We demonstrate how MRC allows AI training jobs to ride out many network failures that previously would have interrupted training.

S. Narayanan, S. Ahmadi, P. Desain et al.

A limitation of brain-computer interface (BCI) spellers is that they require the user to be able to move the eyes to fixate on targets. This poses an issue for users who cannot voluntarily control their eye movements, for instance, people living with late-stage amyotrophic lateral sclerosis (ALS). This pilot study makes the first step towards a gaze-independent speller based on the code-modulated visual evoked potential (c-VEP). Participants were presented with two bi-laterally located stimuli, one of which was flashing, and were tasked to attend to one of these stimuli either by directly looking at the stimuli (overt condition) or by using spatial attention, eliminating the need for eye movement (covert condition). The attended stimuli were decoded from electroencephalography (EEG) and classification accuracies of 88% and 100% were obtained for the covert and overt conditions, respectively. These fundamental insights show the promising feasibility of utilizing the c-VEP protocol for gaze-independent BCIs that use covert spatial attention when both stimuli flash simultaneously.

D. Xezonaki, G. Paraskevopoulos, A. Potamianos et al.

In this work we propose a machine learning model for depression detection from transcribed clinical interviews. Depression is a mental disorder that impacts not only the subject's mood but also the use of language. To this end we use a Hierarchical Attention Network to classify interviews of depressed subjects. We augment the attention layer of our model with a conditioning mechanism on linguistic features, extracted from affective lexica. Our analysis shows that individuals diagnosed with depression use affective language to a greater extent than not-depressed. Our experiments show that external affective information improves the performance of the proposed architecture in the General Psychotherapy Corpus and the DAIC-WoZ 2017 depression datasets, achieving state-of-the-art 71.6 and 68.6 F1 scores respectively.