Transformer-based pre-trained language models boost the performance of open-domain dialogue systems. Prior works leverage Transformer-based pre-trained language models to generate texts with desired attributes in two general approaches: (1) gradient-
based methods: updating all latent representations of pre-trained models with gradients from attribute models; (2) weighted-decoding methods: re-ranking beam candidates from pre-trained models with attribute functions. However, gradient-based methods lead to high computation cost and can easily get overfitted on small training sets, while weighted-decoding methods are inherently constrained by the low-variance high-bias pre-trained model. In this work, we propose a novel approach to control the generation of Transformer-based pre-trained language models: the SideControl framework, which leverages a novel control attributes loss to incorporate useful control signals, and is shown to perform well with very limited training samples. We evaluate our proposed method on two benchmark open-domain dialogue datasets, and results show that the SideControl framework has better controllability, higher generation quality and better sample-efficiency than existing gradient-based and weighted-decoding baselines.
During the three successive seasons of 2002, 2003, and 2004 six populations,
viz.,P1,P2,F1,F2,BC1, and BC2 of two single crosses of cotton (Aleppo40 x
Aleppo90) and (Aleppo33/1 x Line118) were established and evaluated to
explore heterosis, inbree
ding depression, heritability and genetic advance
under selection for reproductive branches number, number of bolls per plant,
boll weight, and ginning percentage. Significant positive heterotic effects were
found for all the studied traits in two crosses except ginning percentage (in the
second cross). Significant positive heterosis over mid-parent and higher parent
for yield components (number of bolls per plant and boll weight) may lead to a
positive heterosis for seed cotton yield/plant.
Succes of a breeding program depends, to a great extent, on understanding
the gentic behavior of the traits. This investigation was conducted to study and
examine the inheritance of some economically important characters such as:
total bolls numbe
r, actual bolls number, ginning percentage , boll weight,
harvest index, flowering earliness, maturity earliness, plant height,and fiber
fineness.
Heritability and genetic advance of some grain yield components and some
morpho-physiological traits in 36 sorghum hybrids [Sorghum bicolor L.
Moench] produced by top-cross using cytoplasmic male sterility were
estimated.
Narrow-sense heritabilit
y for plant height and head length were relatively
high in Damascus location (65% and 56%) respectively, but the narrow-sense
heritability of all characters in Raqqa location were low, the range was from
3% for threshing percentage to 34% for plant height, and from 22% for seed
weight/head to 60% for plant height in the combined analysis.
Genetic advance estimates were relatively low for all characters, the range
was from 0% for 100 seed weight to 28.3% for plant height (Damascus
location), and from 0.02% for seed numbers/head to 15.2% for plant height
(Raqqa location), and from 5.8% for the number of days to flowering to 26.5%
for plant height (combined analysis).
Some characters such as plant height and head length showed relatively
high values of heritability, but exhibited low values of genetic advance.
Therefore other characters had low heritability coupled with low values of
genetic advance indicated predominance of non-additive gene action. This
result implies the need for several generations of selection to improve grain
yield components and other traits.
