Since this is a conference talk rather than a journal paper, I'll start on a philosophical note. Let's reflect on some of the good fortune observational cosmologists enjoy. The task of understanding the contents of the universe, (let alone their evolution), is so audacious, that it is remarkable we can even start doing the job. One reason we can is that gravity helpfully imposes some clear organization on how matter is arranged. Luminous mass appears (at most wavelengths) to be strikingly organized into stars and galaxies. Biologists rely heavily on their crucial organizing concepts of cell, organism and species. As astronomers, we see the world clearly arranged into self-gravitating planets, stars and galaxies. Although these categories span enormous ranges, for example in luminosity, they obey regularities and patterns. This is why we can make clear distinctions between what is or is not a star, and what is or is not a galaxy. We can, for example, distinguish individuals in these classes from bound clusters of them. These are two of the surprisingly easy categories which define our field. When we cannot use them, our job gets harder. Operationally, the star and galaxy concepts are closely linked, since we rarely think about one being present without the other. It is in fact very difficult to learn much about "galaxies" before they began forming many stars. I'll discuss one possible exception to that generalization at the end.
The last half of the 20th century saw dramatic progress in understanding the evolution and even the formation of stars. The first half of the 21st will herald similar advances in understanding the evolution and formation of galaxies. These systems are sufficiently more complicated and distant than stars that empirical, observation- driven data must play a relatively larger role than theory. (Stellar structure and evolution are rarely influenced by external circumstances. Isolated stars, for example, hardly ever experience close encounters. Galaxies often do, and their motions are often dominated by dark matter we know little about.)
Since galaxies have a wide range of (evolving) properties, we need to study large samples of them. Fortunately, they are so numerous that deep field surveys with large array detectors can provide such samples, often spanning a wide range of redshift simultaneously. Two of the most fundamental observables to derive from galaxy surveys are 1) their luminosity function (a measure of their population of formed stars) and 2) their rates of recent and ongoing star formation. In deep fields we can study how both of these evolve through cosmic time.
Of the results obtained from the Hubble Deep Field (HDF),
the most far-reaching is that the star formation
rate is high at z ~ 3
([20],
[21]),
with a possible decline at z ~ 4.
When combined with low-redshift data (e.g.
[18]),
there is evidence for a peak in the comoving SFR at
z 
1 - 2,
around a third of the Hubble Time. The apparent resemblance between
this star formation rate curve and the evolution of quasars
has been noted.
For both of the main goals in the study of galaxy evolution-measuring the stellar population, and its first time derivative-observations with large-format (e.g.million-pixel) infrared detectors are essential. I will illustrate this by reviewing several applications of infrared observations. Some of these were also discussed at a Tokyo conference of the Birth and Evolution of the Universe ([24]). Even though only two years have passed, so many advances have been made, that an update is due.