User profiles for N. Jojic
Nebojsa JojicMicrosoft Research Verified email at microsoft.com Cited by 11471 |
The human splicing code reveals new insights into the genetic determinants of disease
… For quality control, we only examined exon-tissue combinations (n = 56,784) for which the
… in ASD cases (n = 171), 27% have high expression in brain, whereas for controls (n = 249), …
… in ASD cases (n = 171), 27% have high expression in brain, whereas for controls (n = 249), …
Locus: Learning object classes with unsupervised segmentation
We address the problem of learning object class models and object segmentations from
unannotated images. We introduce LOCUS (learning object classes with unsupervised …
unannotated images. We introduce LOCUS (learning object classes with unsupervised …
Diffusion models as plug-and-play priors
… The sample x is initialized with Gaussian noise N(0, I), and as we perform gradient steps
with decreasing values of t, we observe facial features being added in a coarse-to-fine manner. …
with decreasing values of t, we observe facial features being added in a coarse-to-fine manner. …
Learning flexible sprites in video layers
We propose a technique for automatically learning layers of "flexible sprites" (probabilistic 2-dimensional
appearance maps and masks of moving, occluding objects). The model …
appearance maps and masks of moving, occluding objects). The model …
Efficient ranking from pairwise comparisons
… In this paper we show that much simpler algorithms can match the same Ω(n) lower bound
in expectation. Furthermore, if an average of O(n log(n)) binary comparisons are measured, …
in expectation. Furthermore, if an average of O(n log(n)) binary comparisons are measured, …
A comparison of algorithms for inference and learning in probabilistic graphical models
… If n counts the number of sampling steps, then as n ! 1 thenthconfiguration of the hidden RVs
is guaranteed to be an unbiased sample from the exact posterior PðhjvÞ. In fact, although a …
is guaranteed to be an unbiased sample from the exact posterior PðhjvÞ. In fact, although a …
Deep learning of the regulatory grammar of yeast 5′ untranslated regions from 500,000 random sequences
Our ability to predict protein expression from DNA sequence alone remains poor, reflecting
our limited understanding of cis-regulatory grammar and hampering the design of …
our limited understanding of cis-regulatory grammar and hampering the design of …
A deep active learning system for species identification and counting in camera trap images
A typical camera trap survey may produce millions of images that require slow, expensive
manual review. Consequently, critical conservation questions may be answered too slowly to …
manual review. Consequently, critical conservation questions may be answered too slowly to …
Multi-label learning from single positive labels
… We write the estimated labels as Y ∈ [0, 1]N×L in analogy with the matrix of true labels Y ∈
{0, 1}N×L and the matrix of classifier predictions F ∈ [0, 1]N×L. We carry through the derived …
{0, 1}N×L and the matrix of classifier predictions F ∈ [0, 1]N×L. We carry through the derived …
Delayed-release oral mesalamine 4.8 g/day (800-mg tablet) is effective for patients with moderately active ulcerative colitis
…, J Regula, BG Feagan, E Belousova, N Jojic… - Gastroenterology, 2009 - Elsevier
BACKGROUND AND AIMS: It is not clear what induction dose of mesalamine is optimal for
treating patients with mildly and moderately active ulcerative colitis (UC). This study was …
treating patients with mildly and moderately active ulcerative colitis (UC). This study was …